COMPUTER VISION CAPSTONE PROJECT AIML OBJECT DETECTION - CAR¶

AWS Specific code

Installing libraries

In [1]:
!pip install tensorflow opencv-python pillow scikit-learn

Requirement already satisfied: tensorflow in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (2.16.2)
Requirement already satisfied: opencv-python in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (4.11.0.86)
Requirement already satisfied: pillow in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (11.1.0)
Requirement already satisfied: scikit-learn in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (1.6.1)
Requirement already satisfied: absl-py>=1.0.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (2.1.0)
Requirement already satisfied: astunparse>=1.6.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (1.6.3)
Requirement already satisfied: flatbuffers>=23.5.26 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (25.2.10)
Requirement already satisfied: gast!=0.5.0,!=0.5.1,!=0.5.2,>=0.2.1 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (0.6.0)
Requirement already satisfied: google-pasta>=0.1.1 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (0.2.0)
Requirement already satisfied: h5py>=3.10.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (3.12.1)
Requirement already satisfied: libclang>=13.0.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (18.1.1)
Requirement already satisfied: ml-dtypes~=0.3.1 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (0.3.2)
Requirement already satisfied: opt-einsum>=2.3.2 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (3.4.0)
Requirement already satisfied: packaging in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (21.3)
Requirement already satisfied: protobuf!=4.21.0,!=4.21.1,!=4.21.2,!=4.21.3,!=4.21.4,!=4.21.5,<5.0.0dev,>=3.20.3 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (4.25.6)
Requirement already satisfied: requests<3,>=2.21.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (2.32.3)
Requirement already satisfied: setuptools in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (75.8.0)
Requirement already satisfied: six>=1.12.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (1.17.0)
Requirement already satisfied: termcolor>=1.1.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (2.5.0)
Requirement already satisfied: typing-extensions>=3.6.6 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (4.12.2)
Requirement already satisfied: wrapt>=1.11.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (1.17.2)
Requirement already satisfied: grpcio<2.0,>=1.24.3 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (1.70.0)
Requirement already satisfied: tensorboard<2.17,>=2.16 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (2.16.2)
Requirement already satisfied: keras>=3.0.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (3.8.0)
Requirement already satisfied: tensorflow-io-gcs-filesystem>=0.23.1 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (0.37.1)
Requirement already satisfied: numpy<2.0.0,>=1.23.5 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorflow) (1.26.4)
Requirement already satisfied: scipy>=1.6.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from scikit-learn) (1.15.1)
Requirement already satisfied: joblib>=1.2.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from scikit-learn) (1.4.2)
Requirement already satisfied: threadpoolctl>=3.1.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from scikit-learn) (3.5.0)
Requirement already satisfied: wheel<1.0,>=0.23.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from astunparse>=1.6.0->tensorflow) (0.45.1)
Requirement already satisfied: rich in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from keras>=3.0.0->tensorflow) (13.9.4)
Requirement already satisfied: namex in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from keras>=3.0.0->tensorflow) (0.0.8)
Requirement already satisfied: optree in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from keras>=3.0.0->tensorflow) (0.14.0)
Requirement already satisfied: charset_normalizer<4,>=2 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from requests<3,>=2.21.0->tensorflow) (3.4.1)
Requirement already satisfied: idna<4,>=2.5 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from requests<3,>=2.21.0->tensorflow) (3.10)
Requirement already satisfied: urllib3<3,>=1.21.1 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from requests<3,>=2.21.0->tensorflow) (1.26.19)
Requirement already satisfied: certifi>=2017.4.17 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from requests<3,>=2.21.0->tensorflow) (2025.1.31)
Requirement already satisfied: markdown>=2.6.8 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorboard<2.17,>=2.16->tensorflow) (3.7)
Requirement already satisfied: tensorboard-data-server<0.8.0,>=0.7.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorboard<2.17,>=2.16->tensorflow) (0.7.2)
Requirement already satisfied: werkzeug>=1.0.1 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from tensorboard<2.17,>=2.16->tensorflow) (3.1.3)
Requirement already satisfied: pyparsing!=3.0.5,>=2.0.2 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from packaging->tensorflow) (3.2.1)
Requirement already satisfied: MarkupSafe>=2.1.1 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from werkzeug>=1.0.1->tensorboard<2.17,>=2.16->tensorflow) (3.0.2)
Requirement already satisfied: markdown-it-py>=2.2.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from rich->keras>=3.0.0->tensorflow) (3.0.0)
Requirement already satisfied: pygments<3.0.0,>=2.13.0 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from rich->keras>=3.0.0->tensorflow) (2.19.1)
Requirement already satisfied: mdurl~=0.1 in /home/ec2-user/anaconda3/envs/tensorflow2_p310/lib/python3.10/site-packages (from markdown-it-py>=2.2.0->rich->keras>=3.0.0->tensorflow) (0.1.2)
In [2]:
import os
In [3]:
os.environ["TF_ENABLE_ONEDNN_OPTS"] = "0"
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
In [4]:
!nvidia-smi

Sat Mar 29 04:29:17 2025       
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 550.144.03             Driver Version: 550.144.03     CUDA Version: 12.4     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  NVIDIA A10G                    On  |   00000000:00:1E.0 Off |                    0 |
|  0%   30C    P8             16W /  300W |       1MiB /  23028MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
                                                                                         
+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+
In [5]:
import boto3
In [6]:
def download_files_from_bucket(file,bucket):
     '''
     this function is for downloading the files from the bucket to the local instance
     '''    
     bucket_name = bucket
     file_key = file
     local_file_path = file
     s3 = boto3.client('s3')
     s3.download_file(bucket_name, file_key, local_file_path)    
     print(f"File downloaded to {local_file_path}")
In [7]:
download_files_from_bucket('stanford-car-dataset-by-classes-folder.zip','pgp-capstone-project')

File downloaded to stanford-car-dataset-by-classes-folder.zip
In [8]:
zip_file_path = 'stanford-car-dataset-by-classes-folder.zip'
!unzip -oq stanford-car-dataset-by-classes-folder.zip
  1. Problem Statement

Computer vision can be used to automate supervision and generate action appropriate action trigger if the event is predicted from the image of interest. For example a car moving on the road can be easily identified by a camera as make of the car, type, colour, number plates etc.

Design a DL based car identification model.

  1. Introduction

The Cars dataset contains 16,185 images of 196 classes of cars. The data is split into 8,144 training images and 8,041 testing images, where each class has been split roughly in a 50-50 split. Classes are typically at the level of Make, Model, Year, e.g. 2012 Tesla Model S or 2012 BMW M3 coupe.

Data description:

‣ Train Images: Consists of real images of cars as per the make and year of the car.

‣ Test Images: Consists of real images of cars as per the make and year of the car.

‣ Train Annotation: Consists of bounding box region for training images.

‣ Test Annotation: Consists of bounding box region for testing images.

  1. Libraries Used
In [9]:
import os
import zipfile
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt #for visualization
import matplotlib.patches as patches
import seaborn as sns
from PIL import Image # For image loading and manipulation
from pathlib import Path

import cv2
from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report, accuracy_score
from sklearn.metrics import confusion_matrix
from sklearn.preprocessing import LabelEncoder
from sklearn.utils.class_weight import compute_class_weight

import tensorflow as tf
from tensorflow.keras import layers
from tensorflow.keras.models import Sequential, Model
from tensorflow.keras.layers import Conv2D, MaxPooling2D, Flatten, Dense, Dropout, GlobalAveragePooling2D, BatchNormalization
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.preprocessing.image import ImageDataGenerator
from tensorflow.keras.optimizers import Adam
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint
from sklearn.utils import class_weight

from tensorflow.keras.applications.resnet50 import preprocess_input as resnet_preprocess
from tensorflow.keras.applications.mobilenet_v2 import preprocess_input as mobilenet_preprocess
from keras.applications.inception_v3 import preprocess_input as googlenet_preprocess


from tensorflow.keras.applications import MobileNetV2
from tensorflow.keras.applications import InceptionV3
from tensorflow.keras.applications import ResNet50

Matplotlib is building the font cache; this may take a moment.
In [10]:
gpus = tf.config.experimental.list_physical_devices('GPU')
if gpus:
    try:
        for gpu in gpus:
            tf.config.experimental.set_memory_growth(gpu, True)  # Allow dynamic allocation
    except RuntimeError as e:
        print(e)
  1. Data Handling

4A. Data Handling - Import Data

In [11]:
train_annotations_df = pd.read_csv( "anno_train.csv",header=None)
In [12]:
test_annotations_df = pd.read_csv( "anno_test.csv",header=None)
In [13]:
image_class_df = pd.read_csv( "names.csv",header=None)
In [14]:
train_annotations_df.rename(columns={0:"image_name",1:"xmin",2:"ymin",3:'xmax',4:'ymax',5:'image_class'},inplace=True)
In [15]:
test_annotations_df.rename(columns={0:"image_name",1:"xmin",2:"ymin",3:'xmax',4:'ymax',5:'image_class'},inplace=True)
In [16]:
image_class_df.rename(columns={0:'image_name'},inplace=True)
In [17]:
train_annotations_df.head(5)
Out[17]:
image_name xmin ymin xmax ymax image_class
0 00001.jpg 39 116 569 375 14
1 00002.jpg 36 116 868 587 3
2 00003.jpg 85 109 601 381 91
3 00004.jpg 621 393 1484 1096 134
4 00005.jpg 14 36 133 99 106
In [18]:
test_annotations_df.head()
Out[18]:
image_name xmin ymin xmax ymax image_class
0 00001.jpg 30 52 246 147 181
1 00002.jpg 100 19 576 203 103
2 00003.jpg 51 105 968 659 145
3 00004.jpg 67 84 581 407 187
4 00005.jpg 140 151 593 339 185
In [19]:
# for images
base_dir = Path(r"./car_data") #replace the directory accordingly
train_images_path = base_dir / "car_data" / "train"
test_images_path  = base_dir / "car_data" / "test"
train_images_path = Path(train_images_path).resolve()
test_images_path = Path(test_images_path).resolve()
print(f"train image path is {train_images_path}")
print(f"test image path is {test_images_path}")

train image path is /home/ec2-user/SageMaker/car_data/car_data/train
test image path is /home/ec2-user/SageMaker/car_data/car_data/test

4B. Data Handling - Map Images w.r.t Classes

In [20]:
#Train Images class mapping
#Folder where multiple train images are stored
train_class_folders = [f.path for f in os.scandir(train_images_path) if f.is_dir()]
train_image_classes = {} # Dictionary to store training image: class mapping
train_images_path = list(train_images_path.rglob("*.jpg")) 

# Create a dictionary mapping image filenames to class names (parent folder)
train_image_classes = {img_path.name: img_path.parent.name for img_path in train_images_path}

# Define columns for the Training DataFrame
columns_training = ['Image_Path', 'labels']

# Create an empty DataFrame
df_training = pd.DataFrame(columns=columns_training)
df_training = pd.DataFrame(train_images_path, columns=["Image_Path"])
df_training["labels"] = df_training["Image_Path"].apply(lambda x: Path(x).parent.name) 
df_training["Image_Path"] = df_training["Image_Path"].apply(lambda x: str(Path(x).resolve()))
df_training["Image_Path"] = df_training["Image_Path"].astype(str)
print(df_training.head(10))

# --- Print a few mappings to verify ---
print("Sample Training Image to Class Mappings:")
count = 0
for img_name, class_label in list(train_image_classes.items())[:5]:
    print(f"{img_name}: {class_label}")

                                          Image_Path                  labels
0  /home/ec2-user/SageMaker/car_data/car_data/tra...  Infiniti QX56 SUV 2011
1  /home/ec2-user/SageMaker/car_data/car_data/tra...  Infiniti QX56 SUV 2011
2  /home/ec2-user/SageMaker/car_data/car_data/tra...  Infiniti QX56 SUV 2011
3  /home/ec2-user/SageMaker/car_data/car_data/tra...  Infiniti QX56 SUV 2011
4  /home/ec2-user/SageMaker/car_data/car_data/tra...  Infiniti QX56 SUV 2011
5  /home/ec2-user/SageMaker/car_data/car_data/tra...  Infiniti QX56 SUV 2011
6  /home/ec2-user/SageMaker/car_data/car_data/tra...  Infiniti QX56 SUV 2011
7  /home/ec2-user/SageMaker/car_data/car_data/tra...  Infiniti QX56 SUV 2011
8  /home/ec2-user/SageMaker/car_data/car_data/tra...  Infiniti QX56 SUV 2011
9  /home/ec2-user/SageMaker/car_data/car_data/tra...  Infiniti QX56 SUV 2011
Sample Training Image to Class Mappings:
05829.jpg: Infiniti QX56 SUV 2011
04532.jpg: Infiniti QX56 SUV 2011
04524.jpg: Infiniti QX56 SUV 2011
04856.jpg: Infiniti QX56 SUV 2011
02413.jpg: Infiniti QX56 SUV 2011
In [21]:
#Test Images class mapping
#Folder where multiple test images are stored
test_class_folders = [f.path for f in os.scandir(test_images_path) if f.is_dir()]
test_image_classes = {}  # Dictionary to store testing image: class mapping
test_images_path_root = test_images_path.resolve()
test_images_path_list = list(test_images_path_root.rglob("*.jpg"))

# Create a dictionary mapping image filenames to class names (parent folder)
test_image_classes = {img_path.name: img_path.parent.name for img_path in test_images_path_list}

# Define columns for the Testing DataFrame
columns_testing = ['Image_Path', 'labels']

# Create an empty DataFrame
df_testing = pd.DataFrame(columns=columns_testing)      
df_testing = pd.DataFrame(test_images_path_list, columns=["Image_Path"])
df_testing["labels"] = df_testing["Image_Path"].apply(lambda x: Path(x).parent.name)
df_testing["Image_Path"] = df_testing["Image_Path"].apply(lambda x: str(Path(x).resolve()))
df_testing["Image_Path"] = df_testing["Image_Path"].astype(str)

print(df_testing.head(10))

print("Sample Testing Image to Class Mappings:")
count = 0
for img_name, class_label in list(test_image_classes.items())[:5]:
    print(f"{img_name}: {class_label}")

                                          Image_Path                  labels
0  /home/ec2-user/SageMaker/car_data/car_data/tes...  Infiniti QX56 SUV 2011
1  /home/ec2-user/SageMaker/car_data/car_data/tes...  Infiniti QX56 SUV 2011
2  /home/ec2-user/SageMaker/car_data/car_data/tes...  Infiniti QX56 SUV 2011
3  /home/ec2-user/SageMaker/car_data/car_data/tes...  Infiniti QX56 SUV 2011
4  /home/ec2-user/SageMaker/car_data/car_data/tes...  Infiniti QX56 SUV 2011
5  /home/ec2-user/SageMaker/car_data/car_data/tes...  Infiniti QX56 SUV 2011
6  /home/ec2-user/SageMaker/car_data/car_data/tes...  Infiniti QX56 SUV 2011
7  /home/ec2-user/SageMaker/car_data/car_data/tes...  Infiniti QX56 SUV 2011
8  /home/ec2-user/SageMaker/car_data/car_data/tes...  Infiniti QX56 SUV 2011
9  /home/ec2-user/SageMaker/car_data/car_data/tes...  Infiniti QX56 SUV 2011
Sample Testing Image to Class Mappings:
01068.jpg: Infiniti QX56 SUV 2011
02434.jpg: Infiniti QX56 SUV 2011
02499.jpg: Infiniti QX56 SUV 2011
04803.jpg: Infiniti QX56 SUV 2011
00478.jpg: Infiniti QX56 SUV 2011

4C. Data Handling - Map Images w.r.t Annotations

In [22]:
train_annotations_df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 8144 entries, 0 to 8143
Data columns (total 6 columns):
 #   Column       Non-Null Count  Dtype 
---  ------       --------------  ----- 
 0   image_name   8144 non-null   object
 1   xmin         8144 non-null   int64 
 2   ymin         8144 non-null   int64 
 3   xmax         8144 non-null   int64 
 4   ymax         8144 non-null   int64 
 5   image_class  8144 non-null   int64 
dtypes: int64(5), object(1)
memory usage: 381.9+ KB
In [23]:
# ********Definition of the method ********************************
def map_images_to_bboxes(annotations_file):
    image_bboxes = {}
    try:
        for index, row in annotations_file.iterrows():
                image_name = row['image_name']
                x_min = row['xmin']
                y_min = row['ymin']
                x_max = row['xmax']
                y_max = row['ymax']
                image_class = row['image_class']

                image_bboxes[image_name] = (x_min, y_min, x_max, y_max) # Store bbox as tuple

    except FileNotFoundError:
        print(f"Error: Annotation file not found: {annotations_file}")
    except KeyError as e:
        print(f"Error: Column '{e}' not found in CSV file. Check your CSV column names.")
        print("Expected columns (example): filename, xmin, ymin, xmax, ymax") # Example expected columns

    return image_bboxes
In [24]:
#Train images boundry box mapping
train_image_bboxes = map_images_to_bboxes(train_annotations_df)
# --- Print a few mappings to verify for Training images ---
print("\nSample Training Image to Bounding Box Mappings (DF):")
count = 0
for img_name, bbox in train_image_bboxes.items():
    print(f"{img_name}: {bbox}")
    count += 1
    if count > 5: break

Sample Training Image to Bounding Box Mappings (DF):
00001.jpg: (39, 116, 569, 375)
00002.jpg: (36, 116, 868, 587)
00003.jpg: (85, 109, 601, 381)
00004.jpg: (621, 393, 1484, 1096)
00005.jpg: (14, 36, 133, 99)
00006.jpg: (259, 289, 515, 416)
In [25]:
#Test images boundry box mapping
test_image_bboxes = map_images_to_bboxes(test_annotations_df)
# --- Print a few mappings to verify  testing images---
print("\nSample Testing Image to Bounding Box Mappings (DF):")
count = 0
for img_name, bbox in test_image_bboxes.items():
    print(f"{img_name}: {bbox}")
    count += 1
    if count > 5: break

Sample Testing Image to Bounding Box Mappings (DF):
00001.jpg: (30, 52, 246, 147)
00002.jpg: (100, 19, 576, 203)
00003.jpg: (51, 105, 968, 659)
00004.jpg: (67, 84, 581, 407)
00005.jpg: (140, 151, 593, 339)
00006.jpg: (20, 77, 420, 301)
  1. Display Result - bounding box
In [26]:
# Display images with bounding boxes
def display_image_with_bbox(image_path, annotation):
    # Load image
    img = Image.open(image_path)

    # Create plot
    fig, ax = plt.subplots(1)
    ax.imshow(img)

    # Draw bounding box
    x_min = row['xmin']
    y_min = row['ymin']
    x_max = row['xmax']
    y_max = row['ymax']
    image_class = row['image_class']
    bbox = annotation['bbox']
    rect = patches.Rectangle(
        (x_min, y_min),  # (x_min, y_min) -  (bbox[0], bbox[1])
        (x_max - x_min),   # width (x_max - x_min)  - bbox[2] - bbox[0]
        (y_max - y_min),   # height (y_max - y_min) -- bbox[3] - bbox[1]
        linewidth=2,
        edgecolor='r',
        facecolor='none'
    )
    ax.add_patch(rect)

    # Add class label
    plt.text(
        bbox[0], bbox[1] - 10,  # Position of the label
        annotation['image_class'],
        color='red',
        fontsize=12,
        backgroundcolor='white'
    )

    plt.axis('off')
    plt.show()
In [27]:
# Display bounding box for train images
print("For Training Images") 
displayed_image_count = 0  # Initialize a counter to track displayed images

image_paths_details_training=[]
images_paths_details_testing=[]

for index, row in train_annotations_df.iterrows():
    if displayed_image_count >= 5: # Check if we've already displayed two images
        break  # If yes, exit the loop

    image_name = str(row['image_name']).strip()
    image_path = None # Initialize image_path to None

    for class_folder in train_class_folders:
        potential_image_path = os.path.join(class_folder, image_name)
        if os.path.exists(potential_image_path):
            image_path = potential_image_path
            image_paths_details_training.append(potential_image_path)
            break # Image found, no need to check other class folders

    if image_path: # If image_path is found (not None)
        annotation = {
            'bbox': [row['xmin'], row['ymin'], row['xmax'], row['ymax']],
            'image_class' : row['image_class']
        }
        display_image_with_bbox(image_path, annotation)
        displayed_image_count += 1 # Increment the counter

print(f"Displayed {displayed_image_count} training images with bounding boxes.")

For Training Images

Displayed 5 training images with bounding boxes.
In [28]:
# Display bounding box for test images
print("For Testing Images") 
displayed_image_count_test = 0  # Initialize a counter to track displayed images

for index, row in test_annotations_df.iterrows():  # Use test_annotations_df DataFrame
    if displayed_image_count_test >= 5: # Check if we've already displayed two images (adjust number here if you want 5 or more)
        break  # If yes, exit the loop

    image_name_test = str(row['image_name']).strip()
    image_path_test = None # Initialize image_path_test to None

    for class_folder in test_class_folders: # Use test_class_folders
        potential_image_path_test = os.path.join(class_folder, image_name_test)
        if os.path.exists(potential_image_path_test):
            image_path_test = potential_image_path_test # Assigned to image_path_test
            images_paths_details_testing.append(potential_image_path)
            break # Image found, no need to check other class folders

    if image_path_test: # If image_path_test is found (not None)
        annotation_test = {
            'bbox': [row['xmin'], row['ymin'], row['xmax'], row['ymax']],
            'image_class' : row['image_class'] # Assuming 'Image class' column also exists in test_annotations_df (verify!)
        }
        display_image_with_bbox(image_path_test, annotation_test)  # Changed here
        displayed_image_count_test += 1 # Increment the counter

print(f"Displayed {displayed_image_count_test} test images with bounding boxes.") 

For Testing Images

Displayed 5 test images with bounding boxes.
  1. Design Basic CNN Models

The Models designed are:

  1. MobileNetV2
  2. GoogleNet
  3. AlexNet
  4. ResNet
In [29]:
def preprocess_image(image_path, target_size=(224, 224)):
    """
    Load and preprocess an image for CNN input.
    """
    # Check if the image file exists
    if not os.path.exists(image_path):
        print(f"Warning: Image file not found: {image_path}")
        return None  # Or handle the missing image in a way that makes sense for your application

    image = cv2.imread(image_path)  # Load image

    # Check if image loading was successful
    if image is None:
        print(f"Warning: Failed to load image: {image_path}")
        return None  # Or handle the loading error as needed

    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)  # Convert to RGB
    image = cv2.resize(image, target_size)  # Resize to target size
    image = image / 255.0  # Normalize pixel values to [0, 1]
    return image
In [30]:
def custom_generator(df, batch_size, target_size):
    """
    Custom generator for images and labels.
    """
    num_samples = len(df)
    while True:
        for offset in range(0, num_samples, batch_size):
            batch_samples = df.iloc[offset:offset + batch_size]
            images = []
            labels = []
            for _, row in batch_samples.iterrows():
                image = preprocess_image(row['Image_Path'], target_size)
                label = row['label_categorical']
                images.append(image)
                labels.append(label)
            X = np.array(images, dtype=np.float32)
            y = np.array(labels, dtype=np.float32)
            yield X, y
In [31]:
# Apply preprocessing to all images
df_testing['image'] = df_testing['Image_Path'].apply(preprocess_image)
df_training['image'] = df_training['Image_Path'].apply(preprocess_image)

# Check for and handle None values in the 'image' column
df_testing = df_testing.dropna(subset=['image'])  # Remove rows with None in 'image'
df_training = df_training.dropna(subset=['image'])  # Remove rows with None in 'image'
In [32]:
# Encode labels
label_encoder = LabelEncoder()
df_testing['labels_encoded'] = label_encoder.fit_transform(df_testing['labels'])
df_training['labels_encoded'] = label_encoder.fit_transform(df_training['labels'])

# Convert labels to categorical (one-hot encoding)
df_testing['label_categorical'] = df_testing['labels_encoded'].apply(lambda x: to_categorical(x, num_classes=len(test_class_folders)))
df_training['label_categorical'] = df_training['labels_encoded'].apply(lambda x: to_categorical(x, num_classes=len(test_class_folders)))
In [33]:
# Split df_training into training and validation sets
df_train, df_val = train_test_split(df_training, test_size=0.2, random_state=42)
In [34]:
# Create generators
#batch_size = 32
batch_size = 16
train_generator = custom_generator(df_train, batch_size, target_size=(224, 224))
val_generator = custom_generator(df_val, batch_size, target_size=(224, 224))  # Use df_val for validation
# Test generator remains the same
test_generator = custom_generator(df_testing, batch_size, target_size=(224, 224))
In [35]:
# Check training generator
X_batch, y_batch = next(train_generator)
print("Training batch shape:", X_batch.shape, y_batch.shape)

# Check validation generator
X_batch, y_batch = next(val_generator)
print("Validation batch shape:", X_batch.shape, y_batch.shape)

Training batch shape: (16, 224, 224, 3) (16, 196)
Validation batch shape: (16, 224, 224, 3) (16, 196)
In [36]:
#Generate classification report from a Keras/TensorFlow model using GPU-accelerated prediction.
#Assumes df_val['image'] contains pre-loaded images as np.arrays and df_val['label_categorical'] is one-hot encoded.
#returns y_val_pred, y_val_true: Predicted and true label indices

def generate_classification_report_tf_model(
    model, #model 
    df_val, #val data frame
    label_encoder, #label encoder
    preprocess_fn, #preprocess_input
    batch_size=32,
    report_name="model_report.csv"
):
    
    # Convert image and label columns to NumPy arrays
    images = [img for img in df_val['image'] if img is not None]
    labels = [label for label in df_val['label_categorical'] if label is not None]

    images = np.stack(df_val['image'].values).astype(np.float32)
    labels = np.stack(labels)

    # Build tf.data.Dataset
    dataset = tf.data.Dataset.from_tensor_slices((images, labels))
    dataset = dataset.batch(batch_size).prefetch(tf.data.AUTOTUNE)    
    
    # Predict
    preds = model.predict(dataset, verbose=1)
    y_val_pred = np.argmax(preds, axis=1)
    y_val_true = np.argmax(labels, axis=1)

    # Evaluation
    acc = accuracy_score(y_val_true, y_val_pred)
    print(f"Model Accuracy: {acc:.4f}\n")
    print("Classification Report:")    

    # Save as CSV
    report = classification_report(
        y_val_true, y_val_pred,
        target_names=label_encoder.classes_,
        output_dict=True,
        zero_division=1
    )

    df_report = pd.DataFrame(report).transpose()
    df_report.loc["overall_accuracy"] = [acc, None, None, None]
    df_report.to_csv(report_name)    
    print(f"Report saved as: {report_name}")

    # Print only the average metrics
    print(f"Model Accuracy: {acc:.4f}")
    print("Average Summary Metrics:")
    print(df_report.loc[["macro avg", "weighted avg", "overall_accuracy"]][["precision", "recall", "f1-score"]])   
    
    return y_val_pred, y_val_true, df_report
In [37]:
def plot_training_history(history):
    fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(14, 5))

    # Plot accuracy
    ax1.plot(history.history['accuracy'], label='Training Accuracy')
    ax1.plot(history.history['val_accuracy'], label='Validation Accuracy')
    ax1.set_title('Model Accuracy')
    ax1.set_xlabel('Epochs')
    ax1.set_ylabel('Accuracy')
    ax1.legend()

    # Plot loss
    ax2.plot(history.history['loss'], label='Training Loss')
    ax2.plot(history.history['val_loss'], label='Validation Loss')
    ax2.set_title('Model Loss')
    ax2.set_xlabel('Epochs')
    ax2.set_ylabel('Loss')
    ax2.legend()

    plt.show()

6A. MobileNetV2

In [38]:
base_model = MobileNetV2(weights='imagenet', include_top=False, input_shape=(224, 224, 3),classes=196)  # Use 128x128 for speed

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/mobilenet_v2/mobilenet_v2_weights_tf_dim_ordering_tf_kernels_1.0_224_no_top.h5
9406464/9406464 [==============================] - 0s 0us/step
In [39]:
# Freeze all but last 4 layers for efficient training
for layer in base_model.layers[:-4]:  
    layer.trainable = False  

# Add custom classification layers
x = base_model.output
x = GlobalAveragePooling2D()(x)  # Reduces parameters
x = BatchNormalization()(x)  # Stabilizes training
x = Dense(128, activation='relu')(x)
x = Dropout(0.3)(x)  # Dropout for regularization
predictions = Dense(len(label_encoder.classes_), activation='softmax')(x)  # Output layer
In [40]:
#Split 80-20 of train images
df_train_mobilenet, df_val_mobilenet = train_test_split(df_training, test_size=0.2, random_state=42)
In [41]:
mobilenet_batch_size=16
#df_train_mobilenet_gen = custom_generator(df_train_mobilenet,mobilenet_batch_size,target_size=(128,128))
#df_val_mobilenet_gen = custom_generator(df_val_mobilenet,mobilenet_batch_size,target_size=(128,128))
df_train_mobilenet_gen = custom_generator(df_train_mobilenet,mobilenet_batch_size,target_size=(224,224))
df_val_mobilenet_gen = custom_generator(df_val_mobilenet,mobilenet_batch_size,target_size=(224,224))
In [42]:
# Create the model
mobilenet_model = Model(inputs=base_model.input, outputs=predictions)

# Compile the model
mobilenet_model.compile(optimizer=Adam(learning_rate=0.0001), loss='categorical_crossentropy', metrics=['accuracy'])
mobilenet_model.summary()

# Define steps per epoch
steps_per_epoch = np.ceil(len(df_train_mobilenet) / mobilenet_batch_size).astype(int)
validation_steps = np.ceil(len(df_val_mobilenet) / mobilenet_batch_size).astype(int)

y_true = np.array(df_train_mobilenet['labels_encoded'].tolist())

# Compute class weights based on actual class distribution
class_weights = compute_class_weight('balanced', classes=np.unique(y_true), y=y_true)
class_weight_dict = dict(enumerate(class_weights))

#predicting
history_mobilenet = mobilenet_model.fit(
     df_train_mobilenet_gen,
    steps_per_epoch=steps_per_epoch,
    validation_data=df_val_mobilenet_gen,
    validation_steps=validation_steps,
    epochs=10  # Reduce epochs to speed up training
)

Model: "model"
__________________________________________________________________________________________________
 Layer (type)                Output Shape                 Param #   Connected to                  
==================================================================================================
 input_1 (InputLayer)        [(None, 224, 224, 3)]        0         []                            
                                                                                                  
 Conv1 (Conv2D)              (None, 112, 112, 32)         864       ['input_1[0][0]']             
                                                                                                  
 bn_Conv1 (BatchNormalizati  (None, 112, 112, 32)         128       ['Conv1[0][0]']               
 on)                                                                                              
                                                                                                  
 Conv1_relu (ReLU)           (None, 112, 112, 32)         0         ['bn_Conv1[0][0]']            
                                                                                                  
 expanded_conv_depthwise (D  (None, 112, 112, 32)         288       ['Conv1_relu[0][0]']          
 epthwiseConv2D)                                                                                  
                                                                                                  
 expanded_conv_depthwise_BN  (None, 112, 112, 32)         128       ['expanded_conv_depthwise[0][0
  (BatchNormalization)                                              ]']                           
                                                                                                  
 expanded_conv_depthwise_re  (None, 112, 112, 32)         0         ['expanded_conv_depthwise_BN[0
 lu (ReLU)                                                          ][0]']                        
                                                                                                  
 expanded_conv_project (Con  (None, 112, 112, 16)         512       ['expanded_conv_depthwise_relu
 v2D)                                                               [0][0]']                      
                                                                                                  
 expanded_conv_project_BN (  (None, 112, 112, 16)         64        ['expanded_conv_project[0][0]'
 BatchNormalization)                                                ]                             
                                                                                                  
 block_1_expand (Conv2D)     (None, 112, 112, 96)         1536      ['expanded_conv_project_BN[0][
                                                                    0]']                          
                                                                                                  
 block_1_expand_BN (BatchNo  (None, 112, 112, 96)         384       ['block_1_expand[0][0]']      
 rmalization)                                                                                     
                                                                                                  
 block_1_expand_relu (ReLU)  (None, 112, 112, 96)         0         ['block_1_expand_BN[0][0]']   
                                                                                                  
 block_1_pad (ZeroPadding2D  (None, 113, 113, 96)         0         ['block_1_expand_relu[0][0]'] 
 )                                                                                                
                                                                                                  
 block_1_depthwise (Depthwi  (None, 56, 56, 96)           864       ['block_1_pad[0][0]']         
 seConv2D)                                                                                        
                                                                                                  
 block_1_depthwise_BN (Batc  (None, 56, 56, 96)           384       ['block_1_depthwise[0][0]']   
 hNormalization)                                                                                  
                                                                                                  
 block_1_depthwise_relu (Re  (None, 56, 56, 96)           0         ['block_1_depthwise_BN[0][0]']
 LU)                                                                                              
                                                                                                  
 block_1_project (Conv2D)    (None, 56, 56, 24)           2304      ['block_1_depthwise_relu[0][0]
                                                                    ']                            
                                                                                                  
 block_1_project_BN (BatchN  (None, 56, 56, 24)           96        ['block_1_project[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_2_expand (Conv2D)     (None, 56, 56, 144)          3456      ['block_1_project_BN[0][0]']  
                                                                                                  
 block_2_expand_BN (BatchNo  (None, 56, 56, 144)          576       ['block_2_expand[0][0]']      
 rmalization)                                                                                     
                                                                                                  
 block_2_expand_relu (ReLU)  (None, 56, 56, 144)          0         ['block_2_expand_BN[0][0]']   
                                                                                                  
 block_2_depthwise (Depthwi  (None, 56, 56, 144)          1296      ['block_2_expand_relu[0][0]'] 
 seConv2D)                                                                                        
                                                                                                  
 block_2_depthwise_BN (Batc  (None, 56, 56, 144)          576       ['block_2_depthwise[0][0]']   
 hNormalization)                                                                                  
                                                                                                  
 block_2_depthwise_relu (Re  (None, 56, 56, 144)          0         ['block_2_depthwise_BN[0][0]']
 LU)                                                                                              
                                                                                                  
 block_2_project (Conv2D)    (None, 56, 56, 24)           3456      ['block_2_depthwise_relu[0][0]
                                                                    ']                            
                                                                                                  
 block_2_project_BN (BatchN  (None, 56, 56, 24)           96        ['block_2_project[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_2_add (Add)           (None, 56, 56, 24)           0         ['block_1_project_BN[0][0]',  
                                                                     'block_2_project_BN[0][0]']  
                                                                                                  
 block_3_expand (Conv2D)     (None, 56, 56, 144)          3456      ['block_2_add[0][0]']         
                                                                                                  
 block_3_expand_BN (BatchNo  (None, 56, 56, 144)          576       ['block_3_expand[0][0]']      
 rmalization)                                                                                     
                                                                                                  
 block_3_expand_relu (ReLU)  (None, 56, 56, 144)          0         ['block_3_expand_BN[0][0]']   
                                                                                                  
 block_3_pad (ZeroPadding2D  (None, 57, 57, 144)          0         ['block_3_expand_relu[0][0]'] 
 )                                                                                                
                                                                                                  
 block_3_depthwise (Depthwi  (None, 28, 28, 144)          1296      ['block_3_pad[0][0]']         
 seConv2D)                                                                                        
                                                                                                  
 block_3_depthwise_BN (Batc  (None, 28, 28, 144)          576       ['block_3_depthwise[0][0]']   
 hNormalization)                                                                                  
                                                                                                  
 block_3_depthwise_relu (Re  (None, 28, 28, 144)          0         ['block_3_depthwise_BN[0][0]']
 LU)                                                                                              
                                                                                                  
 block_3_project (Conv2D)    (None, 28, 28, 32)           4608      ['block_3_depthwise_relu[0][0]
                                                                    ']                            
                                                                                                  
 block_3_project_BN (BatchN  (None, 28, 28, 32)           128       ['block_3_project[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_4_expand (Conv2D)     (None, 28, 28, 192)          6144      ['block_3_project_BN[0][0]']  
                                                                                                  
 block_4_expand_BN (BatchNo  (None, 28, 28, 192)          768       ['block_4_expand[0][0]']      
 rmalization)                                                                                     
                                                                                                  
 block_4_expand_relu (ReLU)  (None, 28, 28, 192)          0         ['block_4_expand_BN[0][0]']   
                                                                                                  
 block_4_depthwise (Depthwi  (None, 28, 28, 192)          1728      ['block_4_expand_relu[0][0]'] 
 seConv2D)                                                                                        
                                                                                                  
 block_4_depthwise_BN (Batc  (None, 28, 28, 192)          768       ['block_4_depthwise[0][0]']   
 hNormalization)                                                                                  
                                                                                                  
 block_4_depthwise_relu (Re  (None, 28, 28, 192)          0         ['block_4_depthwise_BN[0][0]']
 LU)                                                                                              
                                                                                                  
 block_4_project (Conv2D)    (None, 28, 28, 32)           6144      ['block_4_depthwise_relu[0][0]
                                                                    ']                            
                                                                                                  
 block_4_project_BN (BatchN  (None, 28, 28, 32)           128       ['block_4_project[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_4_add (Add)           (None, 28, 28, 32)           0         ['block_3_project_BN[0][0]',  
                                                                     'block_4_project_BN[0][0]']  
                                                                                                  
 block_5_expand (Conv2D)     (None, 28, 28, 192)          6144      ['block_4_add[0][0]']         
                                                                                                  
 block_5_expand_BN (BatchNo  (None, 28, 28, 192)          768       ['block_5_expand[0][0]']      
 rmalization)                                                                                     
                                                                                                  
 block_5_expand_relu (ReLU)  (None, 28, 28, 192)          0         ['block_5_expand_BN[0][0]']   
                                                                                                  
 block_5_depthwise (Depthwi  (None, 28, 28, 192)          1728      ['block_5_expand_relu[0][0]'] 
 seConv2D)                                                                                        
                                                                                                  
 block_5_depthwise_BN (Batc  (None, 28, 28, 192)          768       ['block_5_depthwise[0][0]']   
 hNormalization)                                                                                  
                                                                                                  
 block_5_depthwise_relu (Re  (None, 28, 28, 192)          0         ['block_5_depthwise_BN[0][0]']
 LU)                                                                                              
                                                                                                  
 block_5_project (Conv2D)    (None, 28, 28, 32)           6144      ['block_5_depthwise_relu[0][0]
                                                                    ']                            
                                                                                                  
 block_5_project_BN (BatchN  (None, 28, 28, 32)           128       ['block_5_project[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_5_add (Add)           (None, 28, 28, 32)           0         ['block_4_add[0][0]',         
                                                                     'block_5_project_BN[0][0]']  
                                                                                                  
 block_6_expand (Conv2D)     (None, 28, 28, 192)          6144      ['block_5_add[0][0]']         
                                                                                                  
 block_6_expand_BN (BatchNo  (None, 28, 28, 192)          768       ['block_6_expand[0][0]']      
 rmalization)                                                                                     
                                                                                                  
 block_6_expand_relu (ReLU)  (None, 28, 28, 192)          0         ['block_6_expand_BN[0][0]']   
                                                                                                  
 block_6_pad (ZeroPadding2D  (None, 29, 29, 192)          0         ['block_6_expand_relu[0][0]'] 
 )                                                                                                
                                                                                                  
 block_6_depthwise (Depthwi  (None, 14, 14, 192)          1728      ['block_6_pad[0][0]']         
 seConv2D)                                                                                        
                                                                                                  
 block_6_depthwise_BN (Batc  (None, 14, 14, 192)          768       ['block_6_depthwise[0][0]']   
 hNormalization)                                                                                  
                                                                                                  
 block_6_depthwise_relu (Re  (None, 14, 14, 192)          0         ['block_6_depthwise_BN[0][0]']
 LU)                                                                                              
                                                                                                  
 block_6_project (Conv2D)    (None, 14, 14, 64)           12288     ['block_6_depthwise_relu[0][0]
                                                                    ']                            
                                                                                                  
 block_6_project_BN (BatchN  (None, 14, 14, 64)           256       ['block_6_project[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_7_expand (Conv2D)     (None, 14, 14, 384)          24576     ['block_6_project_BN[0][0]']  
                                                                                                  
 block_7_expand_BN (BatchNo  (None, 14, 14, 384)          1536      ['block_7_expand[0][0]']      
 rmalization)                                                                                     
                                                                                                  
 block_7_expand_relu (ReLU)  (None, 14, 14, 384)          0         ['block_7_expand_BN[0][0]']   
                                                                                                  
 block_7_depthwise (Depthwi  (None, 14, 14, 384)          3456      ['block_7_expand_relu[0][0]'] 
 seConv2D)                                                                                        
                                                                                                  
 block_7_depthwise_BN (Batc  (None, 14, 14, 384)          1536      ['block_7_depthwise[0][0]']   
 hNormalization)                                                                                  
                                                                                                  
 block_7_depthwise_relu (Re  (None, 14, 14, 384)          0         ['block_7_depthwise_BN[0][0]']
 LU)                                                                                              
                                                                                                  
 block_7_project (Conv2D)    (None, 14, 14, 64)           24576     ['block_7_depthwise_relu[0][0]
                                                                    ']                            
                                                                                                  
 block_7_project_BN (BatchN  (None, 14, 14, 64)           256       ['block_7_project[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_7_add (Add)           (None, 14, 14, 64)           0         ['block_6_project_BN[0][0]',  
                                                                     'block_7_project_BN[0][0]']  
                                                                                                  
 block_8_expand (Conv2D)     (None, 14, 14, 384)          24576     ['block_7_add[0][0]']         
                                                                                                  
 block_8_expand_BN (BatchNo  (None, 14, 14, 384)          1536      ['block_8_expand[0][0]']      
 rmalization)                                                                                     
                                                                                                  
 block_8_expand_relu (ReLU)  (None, 14, 14, 384)          0         ['block_8_expand_BN[0][0]']   
                                                                                                  
 block_8_depthwise (Depthwi  (None, 14, 14, 384)          3456      ['block_8_expand_relu[0][0]'] 
 seConv2D)                                                                                        
                                                                                                  
 block_8_depthwise_BN (Batc  (None, 14, 14, 384)          1536      ['block_8_depthwise[0][0]']   
 hNormalization)                                                                                  
                                                                                                  
 block_8_depthwise_relu (Re  (None, 14, 14, 384)          0         ['block_8_depthwise_BN[0][0]']
 LU)                                                                                              
                                                                                                  
 block_8_project (Conv2D)    (None, 14, 14, 64)           24576     ['block_8_depthwise_relu[0][0]
                                                                    ']                            
                                                                                                  
 block_8_project_BN (BatchN  (None, 14, 14, 64)           256       ['block_8_project[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_8_add (Add)           (None, 14, 14, 64)           0         ['block_7_add[0][0]',         
                                                                     'block_8_project_BN[0][0]']  
                                                                                                  
 block_9_expand (Conv2D)     (None, 14, 14, 384)          24576     ['block_8_add[0][0]']         
                                                                                                  
 block_9_expand_BN (BatchNo  (None, 14, 14, 384)          1536      ['block_9_expand[0][0]']      
 rmalization)                                                                                     
                                                                                                  
 block_9_expand_relu (ReLU)  (None, 14, 14, 384)          0         ['block_9_expand_BN[0][0]']   
                                                                                                  
 block_9_depthwise (Depthwi  (None, 14, 14, 384)          3456      ['block_9_expand_relu[0][0]'] 
 seConv2D)                                                                                        
                                                                                                  
 block_9_depthwise_BN (Batc  (None, 14, 14, 384)          1536      ['block_9_depthwise[0][0]']   
 hNormalization)                                                                                  
                                                                                                  
 block_9_depthwise_relu (Re  (None, 14, 14, 384)          0         ['block_9_depthwise_BN[0][0]']
 LU)                                                                                              
                                                                                                  
 block_9_project (Conv2D)    (None, 14, 14, 64)           24576     ['block_9_depthwise_relu[0][0]
                                                                    ']                            
                                                                                                  
 block_9_project_BN (BatchN  (None, 14, 14, 64)           256       ['block_9_project[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_9_add (Add)           (None, 14, 14, 64)           0         ['block_8_add[0][0]',         
                                                                     'block_9_project_BN[0][0]']  
                                                                                                  
 block_10_expand (Conv2D)    (None, 14, 14, 384)          24576     ['block_9_add[0][0]']         
                                                                                                  
 block_10_expand_BN (BatchN  (None, 14, 14, 384)          1536      ['block_10_expand[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_10_expand_relu (ReLU  (None, 14, 14, 384)          0         ['block_10_expand_BN[0][0]']  
 )                                                                                                
                                                                                                  
 block_10_depthwise (Depthw  (None, 14, 14, 384)          3456      ['block_10_expand_relu[0][0]']
 iseConv2D)                                                                                       
                                                                                                  
 block_10_depthwise_BN (Bat  (None, 14, 14, 384)          1536      ['block_10_depthwise[0][0]']  
 chNormalization)                                                                                 
                                                                                                  
 block_10_depthwise_relu (R  (None, 14, 14, 384)          0         ['block_10_depthwise_BN[0][0]'
 eLU)                                                               ]                             
                                                                                                  
 block_10_project (Conv2D)   (None, 14, 14, 96)           36864     ['block_10_depthwise_relu[0][0
                                                                    ]']                           
                                                                                                  
 block_10_project_BN (Batch  (None, 14, 14, 96)           384       ['block_10_project[0][0]']    
 Normalization)                                                                                   
                                                                                                  
 block_11_expand (Conv2D)    (None, 14, 14, 576)          55296     ['block_10_project_BN[0][0]'] 
                                                                                                  
 block_11_expand_BN (BatchN  (None, 14, 14, 576)          2304      ['block_11_expand[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_11_expand_relu (ReLU  (None, 14, 14, 576)          0         ['block_11_expand_BN[0][0]']  
 )                                                                                                
                                                                                                  
 block_11_depthwise (Depthw  (None, 14, 14, 576)          5184      ['block_11_expand_relu[0][0]']
 iseConv2D)                                                                                       
                                                                                                  
 block_11_depthwise_BN (Bat  (None, 14, 14, 576)          2304      ['block_11_depthwise[0][0]']  
 chNormalization)                                                                                 
                                                                                                  
 block_11_depthwise_relu (R  (None, 14, 14, 576)          0         ['block_11_depthwise_BN[0][0]'
 eLU)                                                               ]                             
                                                                                                  
 block_11_project (Conv2D)   (None, 14, 14, 96)           55296     ['block_11_depthwise_relu[0][0
                                                                    ]']                           
                                                                                                  
 block_11_project_BN (Batch  (None, 14, 14, 96)           384       ['block_11_project[0][0]']    
 Normalization)                                                                                   
                                                                                                  
 block_11_add (Add)          (None, 14, 14, 96)           0         ['block_10_project_BN[0][0]', 
                                                                     'block_11_project_BN[0][0]'] 
                                                                                                  
 block_12_expand (Conv2D)    (None, 14, 14, 576)          55296     ['block_11_add[0][0]']        
                                                                                                  
 block_12_expand_BN (BatchN  (None, 14, 14, 576)          2304      ['block_12_expand[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_12_expand_relu (ReLU  (None, 14, 14, 576)          0         ['block_12_expand_BN[0][0]']  
 )                                                                                                
                                                                                                  
 block_12_depthwise (Depthw  (None, 14, 14, 576)          5184      ['block_12_expand_relu[0][0]']
 iseConv2D)                                                                                       
                                                                                                  
 block_12_depthwise_BN (Bat  (None, 14, 14, 576)          2304      ['block_12_depthwise[0][0]']  
 chNormalization)                                                                                 
                                                                                                  
 block_12_depthwise_relu (R  (None, 14, 14, 576)          0         ['block_12_depthwise_BN[0][0]'
 eLU)                                                               ]                             
                                                                                                  
 block_12_project (Conv2D)   (None, 14, 14, 96)           55296     ['block_12_depthwise_relu[0][0
                                                                    ]']                           
                                                                                                  
 block_12_project_BN (Batch  (None, 14, 14, 96)           384       ['block_12_project[0][0]']    
 Normalization)                                                                                   
                                                                                                  
 block_12_add (Add)          (None, 14, 14, 96)           0         ['block_11_add[0][0]',        
                                                                     'block_12_project_BN[0][0]'] 
                                                                                                  
 block_13_expand (Conv2D)    (None, 14, 14, 576)          55296     ['block_12_add[0][0]']        
                                                                                                  
 block_13_expand_BN (BatchN  (None, 14, 14, 576)          2304      ['block_13_expand[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_13_expand_relu (ReLU  (None, 14, 14, 576)          0         ['block_13_expand_BN[0][0]']  
 )                                                                                                
                                                                                                  
 block_13_pad (ZeroPadding2  (None, 15, 15, 576)          0         ['block_13_expand_relu[0][0]']
 D)                                                                                               
                                                                                                  
 block_13_depthwise (Depthw  (None, 7, 7, 576)            5184      ['block_13_pad[0][0]']        
 iseConv2D)                                                                                       
                                                                                                  
 block_13_depthwise_BN (Bat  (None, 7, 7, 576)            2304      ['block_13_depthwise[0][0]']  
 chNormalization)                                                                                 
                                                                                                  
 block_13_depthwise_relu (R  (None, 7, 7, 576)            0         ['block_13_depthwise_BN[0][0]'
 eLU)                                                               ]                             
                                                                                                  
 block_13_project (Conv2D)   (None, 7, 7, 160)            92160     ['block_13_depthwise_relu[0][0
                                                                    ]']                           
                                                                                                  
 block_13_project_BN (Batch  (None, 7, 7, 160)            640       ['block_13_project[0][0]']    
 Normalization)                                                                                   
                                                                                                  
 block_14_expand (Conv2D)    (None, 7, 7, 960)            153600    ['block_13_project_BN[0][0]'] 
                                                                                                  
 block_14_expand_BN (BatchN  (None, 7, 7, 960)            3840      ['block_14_expand[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_14_expand_relu (ReLU  (None, 7, 7, 960)            0         ['block_14_expand_BN[0][0]']  
 )                                                                                                
                                                                                                  
 block_14_depthwise (Depthw  (None, 7, 7, 960)            8640      ['block_14_expand_relu[0][0]']
 iseConv2D)                                                                                       
                                                                                                  
 block_14_depthwise_BN (Bat  (None, 7, 7, 960)            3840      ['block_14_depthwise[0][0]']  
 chNormalization)                                                                                 
                                                                                                  
 block_14_depthwise_relu (R  (None, 7, 7, 960)            0         ['block_14_depthwise_BN[0][0]'
 eLU)                                                               ]                             
                                                                                                  
 block_14_project (Conv2D)   (None, 7, 7, 160)            153600    ['block_14_depthwise_relu[0][0
                                                                    ]']                           
                                                                                                  
 block_14_project_BN (Batch  (None, 7, 7, 160)            640       ['block_14_project[0][0]']    
 Normalization)                                                                                   
                                                                                                  
 block_14_add (Add)          (None, 7, 7, 160)            0         ['block_13_project_BN[0][0]', 
                                                                     'block_14_project_BN[0][0]'] 
                                                                                                  
 block_15_expand (Conv2D)    (None, 7, 7, 960)            153600    ['block_14_add[0][0]']        
                                                                                                  
 block_15_expand_BN (BatchN  (None, 7, 7, 960)            3840      ['block_15_expand[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_15_expand_relu (ReLU  (None, 7, 7, 960)            0         ['block_15_expand_BN[0][0]']  
 )                                                                                                
                                                                                                  
 block_15_depthwise (Depthw  (None, 7, 7, 960)            8640      ['block_15_expand_relu[0][0]']
 iseConv2D)                                                                                       
                                                                                                  
 block_15_depthwise_BN (Bat  (None, 7, 7, 960)            3840      ['block_15_depthwise[0][0]']  
 chNormalization)                                                                                 
                                                                                                  
 block_15_depthwise_relu (R  (None, 7, 7, 960)            0         ['block_15_depthwise_BN[0][0]'
 eLU)                                                               ]                             
                                                                                                  
 block_15_project (Conv2D)   (None, 7, 7, 160)            153600    ['block_15_depthwise_relu[0][0
                                                                    ]']                           
                                                                                                  
 block_15_project_BN (Batch  (None, 7, 7, 160)            640       ['block_15_project[0][0]']    
 Normalization)                                                                                   
                                                                                                  
 block_15_add (Add)          (None, 7, 7, 160)            0         ['block_14_add[0][0]',        
                                                                     'block_15_project_BN[0][0]'] 
                                                                                                  
 block_16_expand (Conv2D)    (None, 7, 7, 960)            153600    ['block_15_add[0][0]']        
                                                                                                  
 block_16_expand_BN (BatchN  (None, 7, 7, 960)            3840      ['block_16_expand[0][0]']     
 ormalization)                                                                                    
                                                                                                  
 block_16_expand_relu (ReLU  (None, 7, 7, 960)            0         ['block_16_expand_BN[0][0]']  
 )                                                                                                
                                                                                                  
 block_16_depthwise (Depthw  (None, 7, 7, 960)            8640      ['block_16_expand_relu[0][0]']
 iseConv2D)                                                                                       
                                                                                                  
 block_16_depthwise_BN (Bat  (None, 7, 7, 960)            3840      ['block_16_depthwise[0][0]']  
 chNormalization)                                                                                 
                                                                                                  
 block_16_depthwise_relu (R  (None, 7, 7, 960)            0         ['block_16_depthwise_BN[0][0]'
 eLU)                                                               ]                             
                                                                                                  
 block_16_project (Conv2D)   (None, 7, 7, 320)            307200    ['block_16_depthwise_relu[0][0
                                                                    ]']                           
                                                                                                  
 block_16_project_BN (Batch  (None, 7, 7, 320)            1280      ['block_16_project[0][0]']    
 Normalization)                                                                                   
                                                                                                  
 Conv_1 (Conv2D)             (None, 7, 7, 1280)           409600    ['block_16_project_BN[0][0]'] 
                                                                                                  
 Conv_1_bn (BatchNormalizat  (None, 7, 7, 1280)           5120      ['Conv_1[0][0]']              
 ion)                                                                                             
                                                                                                  
 out_relu (ReLU)             (None, 7, 7, 1280)           0         ['Conv_1_bn[0][0]']           
                                                                                                  
 global_average_pooling2d (  (None, 1280)                 0         ['out_relu[0][0]']            
 GlobalAveragePooling2D)                                                                          
                                                                                                  
 batch_normalization (Batch  (None, 1280)                 5120      ['global_average_pooling2d[0][
 Normalization)                                                     0]']                          
                                                                                                  
 dense (Dense)               (None, 128)                  163968    ['batch_normalization[0][0]'] 
                                                                                                  
 dropout (Dropout)           (None, 128)                  0         ['dense[0][0]']               
                                                                                                  
 dense_1 (Dense)             (None, 196)                  25284     ['dropout[0][0]']             
                                                                                                  
==================================================================================================
Total params: 2452356 (9.35 MB)
Trainable params: 604612 (2.31 MB)
Non-trainable params: 1847744 (7.05 MB)
__________________________________________________________________________________________________
Epoch 1/10

WARNING: All log messages before absl::InitializeLog() is called are written to STDERR
I0000 00:00:1743223099.903668    9687 service.cc:145] XLA service 0x7f7744393810 initialized for platform CUDA (this does not guarantee that XLA will be used). Devices:
I0000 00:00:1743223099.903708    9687 service.cc:153]   StreamExecutor device (0): NVIDIA A10G, Compute Capability 8.6
I0000 00:00:1743223100.579529    9687 device_compiler.h:188] Compiled cluster using XLA!  This line is logged at most once for the lifetime of the process.

408/408 [==============================] - 104s 82ms/step - loss: 5.2777 - accuracy: 0.0227 - val_loss: 4.7975 - val_accuracy: 0.0552
Epoch 2/10
408/408 [==============================] - 27s 65ms/step - loss: 4.2979 - accuracy: 0.1101 - val_loss: 4.2909 - val_accuracy: 0.1130
Epoch 3/10
408/408 [==============================] - 26s 65ms/step - loss: 3.6382 - accuracy: 0.2106 - val_loss: 3.9245 - val_accuracy: 0.1627
Epoch 4/10
408/408 [==============================] - 26s 64ms/step - loss: 3.1051 - accuracy: 0.3119 - val_loss: 3.6571 - val_accuracy: 0.1995
Epoch 5/10
408/408 [==============================] - 25s 62ms/step - loss: 2.6730 - accuracy: 0.4049 - val_loss: 3.4491 - val_accuracy: 0.2314
Epoch 6/10
408/408 [==============================] - 25s 61ms/step - loss: 2.2819 - accuracy: 0.4973 - val_loss: 3.2925 - val_accuracy: 0.2480
Epoch 7/10
408/408 [==============================] - 23s 57ms/step - loss: 1.9640 - accuracy: 0.5695 - val_loss: 3.1616 - val_accuracy: 0.2701
Epoch 8/10
408/408 [==============================] - 21s 52ms/step - loss: 1.6769 - accuracy: 0.6431 - val_loss: 3.0760 - val_accuracy: 0.2897
Epoch 9/10
408/408 [==============================] - 21s 52ms/step - loss: 1.4317 - accuracy: 0.7002 - val_loss: 2.9941 - val_accuracy: 0.3082
Epoch 10/10
408/408 [==============================] - 21s 52ms/step - loss: 1.2151 - accuracy: 0.7527 - val_loss: 2.9341 - val_accuracy: 0.3076
In [43]:
#display model accurance and loss
plot_training_history(history_mobilenet)
In [44]:
y_pred, y_true,df_mobilenet_classification_report = generate_classification_report_tf_model(
    model=mobilenet_model,
    df_val=df_val_mobilenet,
    label_encoder=label_encoder,
    preprocess_fn=mobilenet_preprocess,
    batch_size=32,
    report_name="mobilenet_classification_report.csv"
)

51/51 [==============================] - 5s 20ms/step
Model Accuracy: 0.3069

Classification Report:
Report saved as: mobilenet_classification_report.csv
Model Accuracy: 0.3069
Average Summary Metrics:
                  precision    recall  f1-score
macro avg          0.320829  0.309102  0.289196
weighted avg       0.349378  0.306937  0.301969
overall_accuracy   0.306937       NaN       NaN

Displaying only top 10 class names

In [45]:
df_support = df_mobilenet_classification_report.iloc[:-3]  # exclude average rows
top_10_classes = df_support.sort_values("support", ascending=False).head(10).index.tolist()
In [46]:
top_10_indices = [np.where(label_encoder.classes_ == cls)[0][0] for cls in top_10_classes]
In [47]:
mobilenet_cm = confusion_matrix(y_true, y_pred)
In [48]:
mobilenet_cm_top10 = mobilenet_cm[np.ix_(top_10_indices, top_10_indices)]
In [49]:
plt.figure(figsize=(10, 8))
sns.heatmap(mobilenet_cm_top10, annot=True, fmt='d',
            xticklabels=top_10_classes, yticklabels=top_10_classes,
            cmap='Blues')
plt.title("Mobile Net Confusion Matrix (Top 10 Classes)")
plt.xlabel("Predicted")
plt.ylabel("True")
plt.tight_layout()
plt.show()
MobileNet Model Summary¶
  • Poor Generalization of The Model
  • Based on the Classification report most cases are not predicted correctly.
  • Indications of Database Imbalance is present
  • lightweight model for real-time applications on mobile devices but here car classification and labeling needed

Next Steps

  • handle data imbalance
  • parameter tuning for to enhance performance

6B. GoogleNet

In [50]:
base_model = InceptionV3(weights='imagenet', include_top=False, input_shape=(224, 224, 3))

# Add custom layers on top of the base model
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(1024, activation='relu')(x)
predictions = Dense(196, activation='softmax')(x)

# Define the complete model
googlenet_model = Model(inputs=base_model.input, outputs=predictions)

# Freeze the layers of the base model
for layer in base_model.layers:
    layer.trainable = False

# Compile the model
googlenet_model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/inception_v3/inception_v3_weights_tf_dim_ordering_tf_kernels_notop.h5
87910968/87910968 [==============================] - 7s 0us/step
In [51]:
googlenet_model.summary()

Model: "model_1"
__________________________________________________________________________________________________
 Layer (type)                Output Shape                 Param #   Connected to                  
==================================================================================================
 input_2 (InputLayer)        [(None, 224, 224, 3)]        0         []                            
                                                                                                  
 conv2d (Conv2D)             (None, 111, 111, 32)         864       ['input_2[0][0]']             
                                                                                                  
 batch_normalization_1 (Bat  (None, 111, 111, 32)         96        ['conv2d[0][0]']              
 chNormalization)                                                                                 
                                                                                                  
 activation (Activation)     (None, 111, 111, 32)         0         ['batch_normalization_1[0][0]'
                                                                    ]                             
                                                                                                  
 conv2d_1 (Conv2D)           (None, 109, 109, 32)         9216      ['activation[0][0]']          
                                                                                                  
 batch_normalization_2 (Bat  (None, 109, 109, 32)         96        ['conv2d_1[0][0]']            
 chNormalization)                                                                                 
                                                                                                  
 activation_1 (Activation)   (None, 109, 109, 32)         0         ['batch_normalization_2[0][0]'
                                                                    ]                             
                                                                                                  
 conv2d_2 (Conv2D)           (None, 109, 109, 64)         18432     ['activation_1[0][0]']        
                                                                                                  
 batch_normalization_3 (Bat  (None, 109, 109, 64)         192       ['conv2d_2[0][0]']            
 chNormalization)                                                                                 
                                                                                                  
 activation_2 (Activation)   (None, 109, 109, 64)         0         ['batch_normalization_3[0][0]'
                                                                    ]                             
                                                                                                  
 max_pooling2d (MaxPooling2  (None, 54, 54, 64)           0         ['activation_2[0][0]']        
 D)                                                                                               
                                                                                                  
 conv2d_3 (Conv2D)           (None, 54, 54, 80)           5120      ['max_pooling2d[0][0]']       
                                                                                                  
 batch_normalization_4 (Bat  (None, 54, 54, 80)           240       ['conv2d_3[0][0]']            
 chNormalization)                                                                                 
                                                                                                  
 activation_3 (Activation)   (None, 54, 54, 80)           0         ['batch_normalization_4[0][0]'
                                                                    ]                             
                                                                                                  
 conv2d_4 (Conv2D)           (None, 52, 52, 192)          138240    ['activation_3[0][0]']        
                                                                                                  
 batch_normalization_5 (Bat  (None, 52, 52, 192)          576       ['conv2d_4[0][0]']            
 chNormalization)                                                                                 
                                                                                                  
 activation_4 (Activation)   (None, 52, 52, 192)          0         ['batch_normalization_5[0][0]'
                                                                    ]                             
                                                                                                  
 max_pooling2d_1 (MaxPoolin  (None, 25, 25, 192)          0         ['activation_4[0][0]']        
 g2D)                                                                                             
                                                                                                  
 conv2d_8 (Conv2D)           (None, 25, 25, 64)           12288     ['max_pooling2d_1[0][0]']     
                                                                                                  
 batch_normalization_9 (Bat  (None, 25, 25, 64)           192       ['conv2d_8[0][0]']            
 chNormalization)                                                                                 
                                                                                                  
 activation_8 (Activation)   (None, 25, 25, 64)           0         ['batch_normalization_9[0][0]'
                                                                    ]                             
                                                                                                  
 conv2d_6 (Conv2D)           (None, 25, 25, 48)           9216      ['max_pooling2d_1[0][0]']     
                                                                                                  
 conv2d_9 (Conv2D)           (None, 25, 25, 96)           55296     ['activation_8[0][0]']        
                                                                                                  
 batch_normalization_7 (Bat  (None, 25, 25, 48)           144       ['conv2d_6[0][0]']            
 chNormalization)                                                                                 
                                                                                                  
 batch_normalization_10 (Ba  (None, 25, 25, 96)           288       ['conv2d_9[0][0]']            
 tchNormalization)                                                                                
                                                                                                  
 activation_6 (Activation)   (None, 25, 25, 48)           0         ['batch_normalization_7[0][0]'
                                                                    ]                             
                                                                                                  
 activation_9 (Activation)   (None, 25, 25, 96)           0         ['batch_normalization_10[0][0]
                                                                    ']                            
                                                                                                  
 average_pooling2d (Average  (None, 25, 25, 192)          0         ['max_pooling2d_1[0][0]']     
 Pooling2D)                                                                                       
                                                                                                  
 conv2d_5 (Conv2D)           (None, 25, 25, 64)           12288     ['max_pooling2d_1[0][0]']     
                                                                                                  
 conv2d_7 (Conv2D)           (None, 25, 25, 64)           76800     ['activation_6[0][0]']        
                                                                                                  
 conv2d_10 (Conv2D)          (None, 25, 25, 96)           82944     ['activation_9[0][0]']        
                                                                                                  
 conv2d_11 (Conv2D)          (None, 25, 25, 32)           6144      ['average_pooling2d[0][0]']   
                                                                                                  
 batch_normalization_6 (Bat  (None, 25, 25, 64)           192       ['conv2d_5[0][0]']            
 chNormalization)                                                                                 
                                                                                                  
 batch_normalization_8 (Bat  (None, 25, 25, 64)           192       ['conv2d_7[0][0]']            
 chNormalization)                                                                                 
                                                                                                  
 batch_normalization_11 (Ba  (None, 25, 25, 96)           288       ['conv2d_10[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_12 (Ba  (None, 25, 25, 32)           96        ['conv2d_11[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_5 (Activation)   (None, 25, 25, 64)           0         ['batch_normalization_6[0][0]'
                                                                    ]                             
                                                                                                  
 activation_7 (Activation)   (None, 25, 25, 64)           0         ['batch_normalization_8[0][0]'
                                                                    ]                             
                                                                                                  
 activation_10 (Activation)  (None, 25, 25, 96)           0         ['batch_normalization_11[0][0]
                                                                    ']                            
                                                                                                  
 activation_11 (Activation)  (None, 25, 25, 32)           0         ['batch_normalization_12[0][0]
                                                                    ']                            
                                                                                                  
 mixed0 (Concatenate)        (None, 25, 25, 256)          0         ['activation_5[0][0]',        
                                                                     'activation_7[0][0]',        
                                                                     'activation_10[0][0]',       
                                                                     'activation_11[0][0]']       
                                                                                                  
 conv2d_15 (Conv2D)          (None, 25, 25, 64)           16384     ['mixed0[0][0]']              
                                                                                                  
 batch_normalization_16 (Ba  (None, 25, 25, 64)           192       ['conv2d_15[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_15 (Activation)  (None, 25, 25, 64)           0         ['batch_normalization_16[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_13 (Conv2D)          (None, 25, 25, 48)           12288     ['mixed0[0][0]']              
                                                                                                  
 conv2d_16 (Conv2D)          (None, 25, 25, 96)           55296     ['activation_15[0][0]']       
                                                                                                  
 batch_normalization_14 (Ba  (None, 25, 25, 48)           144       ['conv2d_13[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_17 (Ba  (None, 25, 25, 96)           288       ['conv2d_16[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_13 (Activation)  (None, 25, 25, 48)           0         ['batch_normalization_14[0][0]
                                                                    ']                            
                                                                                                  
 activation_16 (Activation)  (None, 25, 25, 96)           0         ['batch_normalization_17[0][0]
                                                                    ']                            
                                                                                                  
 average_pooling2d_1 (Avera  (None, 25, 25, 256)          0         ['mixed0[0][0]']              
 gePooling2D)                                                                                     
                                                                                                  
 conv2d_12 (Conv2D)          (None, 25, 25, 64)           16384     ['mixed0[0][0]']              
                                                                                                  
 conv2d_14 (Conv2D)          (None, 25, 25, 64)           76800     ['activation_13[0][0]']       
                                                                                                  
 conv2d_17 (Conv2D)          (None, 25, 25, 96)           82944     ['activation_16[0][0]']       
                                                                                                  
 conv2d_18 (Conv2D)          (None, 25, 25, 64)           16384     ['average_pooling2d_1[0][0]'] 
                                                                                                  
 batch_normalization_13 (Ba  (None, 25, 25, 64)           192       ['conv2d_12[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_15 (Ba  (None, 25, 25, 64)           192       ['conv2d_14[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_18 (Ba  (None, 25, 25, 96)           288       ['conv2d_17[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_19 (Ba  (None, 25, 25, 64)           192       ['conv2d_18[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_12 (Activation)  (None, 25, 25, 64)           0         ['batch_normalization_13[0][0]
                                                                    ']                            
                                                                                                  
 activation_14 (Activation)  (None, 25, 25, 64)           0         ['batch_normalization_15[0][0]
                                                                    ']                            
                                                                                                  
 activation_17 (Activation)  (None, 25, 25, 96)           0         ['batch_normalization_18[0][0]
                                                                    ']                            
                                                                                                  
 activation_18 (Activation)  (None, 25, 25, 64)           0         ['batch_normalization_19[0][0]
                                                                    ']                            
                                                                                                  
 mixed1 (Concatenate)        (None, 25, 25, 288)          0         ['activation_12[0][0]',       
                                                                     'activation_14[0][0]',       
                                                                     'activation_17[0][0]',       
                                                                     'activation_18[0][0]']       
                                                                                                  
 conv2d_22 (Conv2D)          (None, 25, 25, 64)           18432     ['mixed1[0][0]']              
                                                                                                  
 batch_normalization_23 (Ba  (None, 25, 25, 64)           192       ['conv2d_22[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_22 (Activation)  (None, 25, 25, 64)           0         ['batch_normalization_23[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_20 (Conv2D)          (None, 25, 25, 48)           13824     ['mixed1[0][0]']              
                                                                                                  
 conv2d_23 (Conv2D)          (None, 25, 25, 96)           55296     ['activation_22[0][0]']       
                                                                                                  
 batch_normalization_21 (Ba  (None, 25, 25, 48)           144       ['conv2d_20[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_24 (Ba  (None, 25, 25, 96)           288       ['conv2d_23[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_20 (Activation)  (None, 25, 25, 48)           0         ['batch_normalization_21[0][0]
                                                                    ']                            
                                                                                                  
 activation_23 (Activation)  (None, 25, 25, 96)           0         ['batch_normalization_24[0][0]
                                                                    ']                            
                                                                                                  
 average_pooling2d_2 (Avera  (None, 25, 25, 288)          0         ['mixed1[0][0]']              
 gePooling2D)                                                                                     
                                                                                                  
 conv2d_19 (Conv2D)          (None, 25, 25, 64)           18432     ['mixed1[0][0]']              
                                                                                                  
 conv2d_21 (Conv2D)          (None, 25, 25, 64)           76800     ['activation_20[0][0]']       
                                                                                                  
 conv2d_24 (Conv2D)          (None, 25, 25, 96)           82944     ['activation_23[0][0]']       
                                                                                                  
 conv2d_25 (Conv2D)          (None, 25, 25, 64)           18432     ['average_pooling2d_2[0][0]'] 
                                                                                                  
 batch_normalization_20 (Ba  (None, 25, 25, 64)           192       ['conv2d_19[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_22 (Ba  (None, 25, 25, 64)           192       ['conv2d_21[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_25 (Ba  (None, 25, 25, 96)           288       ['conv2d_24[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_26 (Ba  (None, 25, 25, 64)           192       ['conv2d_25[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_19 (Activation)  (None, 25, 25, 64)           0         ['batch_normalization_20[0][0]
                                                                    ']                            
                                                                                                  
 activation_21 (Activation)  (None, 25, 25, 64)           0         ['batch_normalization_22[0][0]
                                                                    ']                            
                                                                                                  
 activation_24 (Activation)  (None, 25, 25, 96)           0         ['batch_normalization_25[0][0]
                                                                    ']                            
                                                                                                  
 activation_25 (Activation)  (None, 25, 25, 64)           0         ['batch_normalization_26[0][0]
                                                                    ']                            
                                                                                                  
 mixed2 (Concatenate)        (None, 25, 25, 288)          0         ['activation_19[0][0]',       
                                                                     'activation_21[0][0]',       
                                                                     'activation_24[0][0]',       
                                                                     'activation_25[0][0]']       
                                                                                                  
 conv2d_27 (Conv2D)          (None, 25, 25, 64)           18432     ['mixed2[0][0]']              
                                                                                                  
 batch_normalization_28 (Ba  (None, 25, 25, 64)           192       ['conv2d_27[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_27 (Activation)  (None, 25, 25, 64)           0         ['batch_normalization_28[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_28 (Conv2D)          (None, 25, 25, 96)           55296     ['activation_27[0][0]']       
                                                                                                  
 batch_normalization_29 (Ba  (None, 25, 25, 96)           288       ['conv2d_28[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_28 (Activation)  (None, 25, 25, 96)           0         ['batch_normalization_29[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_26 (Conv2D)          (None, 12, 12, 384)          995328    ['mixed2[0][0]']              
                                                                                                  
 conv2d_29 (Conv2D)          (None, 12, 12, 96)           82944     ['activation_28[0][0]']       
                                                                                                  
 batch_normalization_27 (Ba  (None, 12, 12, 384)          1152      ['conv2d_26[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_30 (Ba  (None, 12, 12, 96)           288       ['conv2d_29[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_26 (Activation)  (None, 12, 12, 384)          0         ['batch_normalization_27[0][0]
                                                                    ']                            
                                                                                                  
 activation_29 (Activation)  (None, 12, 12, 96)           0         ['batch_normalization_30[0][0]
                                                                    ']                            
                                                                                                  
 max_pooling2d_2 (MaxPoolin  (None, 12, 12, 288)          0         ['mixed2[0][0]']              
 g2D)                                                                                             
                                                                                                  
 mixed3 (Concatenate)        (None, 12, 12, 768)          0         ['activation_26[0][0]',       
                                                                     'activation_29[0][0]',       
                                                                     'max_pooling2d_2[0][0]']     
                                                                                                  
 conv2d_34 (Conv2D)          (None, 12, 12, 128)          98304     ['mixed3[0][0]']              
                                                                                                  
 batch_normalization_35 (Ba  (None, 12, 12, 128)          384       ['conv2d_34[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_34 (Activation)  (None, 12, 12, 128)          0         ['batch_normalization_35[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_35 (Conv2D)          (None, 12, 12, 128)          114688    ['activation_34[0][0]']       
                                                                                                  
 batch_normalization_36 (Ba  (None, 12, 12, 128)          384       ['conv2d_35[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_35 (Activation)  (None, 12, 12, 128)          0         ['batch_normalization_36[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_31 (Conv2D)          (None, 12, 12, 128)          98304     ['mixed3[0][0]']              
                                                                                                  
 conv2d_36 (Conv2D)          (None, 12, 12, 128)          114688    ['activation_35[0][0]']       
                                                                                                  
 batch_normalization_32 (Ba  (None, 12, 12, 128)          384       ['conv2d_31[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_37 (Ba  (None, 12, 12, 128)          384       ['conv2d_36[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_31 (Activation)  (None, 12, 12, 128)          0         ['batch_normalization_32[0][0]
                                                                    ']                            
                                                                                                  
 activation_36 (Activation)  (None, 12, 12, 128)          0         ['batch_normalization_37[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_32 (Conv2D)          (None, 12, 12, 128)          114688    ['activation_31[0][0]']       
                                                                                                  
 conv2d_37 (Conv2D)          (None, 12, 12, 128)          114688    ['activation_36[0][0]']       
                                                                                                  
 batch_normalization_33 (Ba  (None, 12, 12, 128)          384       ['conv2d_32[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_38 (Ba  (None, 12, 12, 128)          384       ['conv2d_37[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_32 (Activation)  (None, 12, 12, 128)          0         ['batch_normalization_33[0][0]
                                                                    ']                            
                                                                                                  
 activation_37 (Activation)  (None, 12, 12, 128)          0         ['batch_normalization_38[0][0]
                                                                    ']                            
                                                                                                  
 average_pooling2d_3 (Avera  (None, 12, 12, 768)          0         ['mixed3[0][0]']              
 gePooling2D)                                                                                     
                                                                                                  
 conv2d_30 (Conv2D)          (None, 12, 12, 192)          147456    ['mixed3[0][0]']              
                                                                                                  
 conv2d_33 (Conv2D)          (None, 12, 12, 192)          172032    ['activation_32[0][0]']       
                                                                                                  
 conv2d_38 (Conv2D)          (None, 12, 12, 192)          172032    ['activation_37[0][0]']       
                                                                                                  
 conv2d_39 (Conv2D)          (None, 12, 12, 192)          147456    ['average_pooling2d_3[0][0]'] 
                                                                                                  
 batch_normalization_31 (Ba  (None, 12, 12, 192)          576       ['conv2d_30[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_34 (Ba  (None, 12, 12, 192)          576       ['conv2d_33[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_39 (Ba  (None, 12, 12, 192)          576       ['conv2d_38[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_40 (Ba  (None, 12, 12, 192)          576       ['conv2d_39[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_30 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_31[0][0]
                                                                    ']                            
                                                                                                  
 activation_33 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_34[0][0]
                                                                    ']                            
                                                                                                  
 activation_38 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_39[0][0]
                                                                    ']                            
                                                                                                  
 activation_39 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_40[0][0]
                                                                    ']                            
                                                                                                  
 mixed4 (Concatenate)        (None, 12, 12, 768)          0         ['activation_30[0][0]',       
                                                                     'activation_33[0][0]',       
                                                                     'activation_38[0][0]',       
                                                                     'activation_39[0][0]']       
                                                                                                  
 conv2d_44 (Conv2D)          (None, 12, 12, 160)          122880    ['mixed4[0][0]']              
                                                                                                  
 batch_normalization_45 (Ba  (None, 12, 12, 160)          480       ['conv2d_44[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_44 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_45[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_45 (Conv2D)          (None, 12, 12, 160)          179200    ['activation_44[0][0]']       
                                                                                                  
 batch_normalization_46 (Ba  (None, 12, 12, 160)          480       ['conv2d_45[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_45 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_46[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_41 (Conv2D)          (None, 12, 12, 160)          122880    ['mixed4[0][0]']              
                                                                                                  
 conv2d_46 (Conv2D)          (None, 12, 12, 160)          179200    ['activation_45[0][0]']       
                                                                                                  
 batch_normalization_42 (Ba  (None, 12, 12, 160)          480       ['conv2d_41[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_47 (Ba  (None, 12, 12, 160)          480       ['conv2d_46[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_41 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_42[0][0]
                                                                    ']                            
                                                                                                  
 activation_46 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_47[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_42 (Conv2D)          (None, 12, 12, 160)          179200    ['activation_41[0][0]']       
                                                                                                  
 conv2d_47 (Conv2D)          (None, 12, 12, 160)          179200    ['activation_46[0][0]']       
                                                                                                  
 batch_normalization_43 (Ba  (None, 12, 12, 160)          480       ['conv2d_42[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_48 (Ba  (None, 12, 12, 160)          480       ['conv2d_47[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_42 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_43[0][0]
                                                                    ']                            
                                                                                                  
 activation_47 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_48[0][0]
                                                                    ']                            
                                                                                                  
 average_pooling2d_4 (Avera  (None, 12, 12, 768)          0         ['mixed4[0][0]']              
 gePooling2D)                                                                                     
                                                                                                  
 conv2d_40 (Conv2D)          (None, 12, 12, 192)          147456    ['mixed4[0][0]']              
                                                                                                  
 conv2d_43 (Conv2D)          (None, 12, 12, 192)          215040    ['activation_42[0][0]']       
                                                                                                  
 conv2d_48 (Conv2D)          (None, 12, 12, 192)          215040    ['activation_47[0][0]']       
                                                                                                  
 conv2d_49 (Conv2D)          (None, 12, 12, 192)          147456    ['average_pooling2d_4[0][0]'] 
                                                                                                  
 batch_normalization_41 (Ba  (None, 12, 12, 192)          576       ['conv2d_40[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_44 (Ba  (None, 12, 12, 192)          576       ['conv2d_43[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_49 (Ba  (None, 12, 12, 192)          576       ['conv2d_48[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_50 (Ba  (None, 12, 12, 192)          576       ['conv2d_49[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_40 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_41[0][0]
                                                                    ']                            
                                                                                                  
 activation_43 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_44[0][0]
                                                                    ']                            
                                                                                                  
 activation_48 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_49[0][0]
                                                                    ']                            
                                                                                                  
 activation_49 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_50[0][0]
                                                                    ']                            
                                                                                                  
 mixed5 (Concatenate)        (None, 12, 12, 768)          0         ['activation_40[0][0]',       
                                                                     'activation_43[0][0]',       
                                                                     'activation_48[0][0]',       
                                                                     'activation_49[0][0]']       
                                                                                                  
 conv2d_54 (Conv2D)          (None, 12, 12, 160)          122880    ['mixed5[0][0]']              
                                                                                                  
 batch_normalization_55 (Ba  (None, 12, 12, 160)          480       ['conv2d_54[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_54 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_55[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_55 (Conv2D)          (None, 12, 12, 160)          179200    ['activation_54[0][0]']       
                                                                                                  
 batch_normalization_56 (Ba  (None, 12, 12, 160)          480       ['conv2d_55[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_55 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_56[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_51 (Conv2D)          (None, 12, 12, 160)          122880    ['mixed5[0][0]']              
                                                                                                  
 conv2d_56 (Conv2D)          (None, 12, 12, 160)          179200    ['activation_55[0][0]']       
                                                                                                  
 batch_normalization_52 (Ba  (None, 12, 12, 160)          480       ['conv2d_51[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_57 (Ba  (None, 12, 12, 160)          480       ['conv2d_56[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_51 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_52[0][0]
                                                                    ']                            
                                                                                                  
 activation_56 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_57[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_52 (Conv2D)          (None, 12, 12, 160)          179200    ['activation_51[0][0]']       
                                                                                                  
 conv2d_57 (Conv2D)          (None, 12, 12, 160)          179200    ['activation_56[0][0]']       
                                                                                                  
 batch_normalization_53 (Ba  (None, 12, 12, 160)          480       ['conv2d_52[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_58 (Ba  (None, 12, 12, 160)          480       ['conv2d_57[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_52 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_53[0][0]
                                                                    ']                            
                                                                                                  
 activation_57 (Activation)  (None, 12, 12, 160)          0         ['batch_normalization_58[0][0]
                                                                    ']                            
                                                                                                  
 average_pooling2d_5 (Avera  (None, 12, 12, 768)          0         ['mixed5[0][0]']              
 gePooling2D)                                                                                     
                                                                                                  
 conv2d_50 (Conv2D)          (None, 12, 12, 192)          147456    ['mixed5[0][0]']              
                                                                                                  
 conv2d_53 (Conv2D)          (None, 12, 12, 192)          215040    ['activation_52[0][0]']       
                                                                                                  
 conv2d_58 (Conv2D)          (None, 12, 12, 192)          215040    ['activation_57[0][0]']       
                                                                                                  
 conv2d_59 (Conv2D)          (None, 12, 12, 192)          147456    ['average_pooling2d_5[0][0]'] 
                                                                                                  
 batch_normalization_51 (Ba  (None, 12, 12, 192)          576       ['conv2d_50[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_54 (Ba  (None, 12, 12, 192)          576       ['conv2d_53[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_59 (Ba  (None, 12, 12, 192)          576       ['conv2d_58[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_60 (Ba  (None, 12, 12, 192)          576       ['conv2d_59[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_50 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_51[0][0]
                                                                    ']                            
                                                                                                  
 activation_53 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_54[0][0]
                                                                    ']                            
                                                                                                  
 activation_58 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_59[0][0]
                                                                    ']                            
                                                                                                  
 activation_59 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_60[0][0]
                                                                    ']                            
                                                                                                  
 mixed6 (Concatenate)        (None, 12, 12, 768)          0         ['activation_50[0][0]',       
                                                                     'activation_53[0][0]',       
                                                                     'activation_58[0][0]',       
                                                                     'activation_59[0][0]']       
                                                                                                  
 conv2d_64 (Conv2D)          (None, 12, 12, 192)          147456    ['mixed6[0][0]']              
                                                                                                  
 batch_normalization_65 (Ba  (None, 12, 12, 192)          576       ['conv2d_64[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_64 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_65[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_65 (Conv2D)          (None, 12, 12, 192)          258048    ['activation_64[0][0]']       
                                                                                                  
 batch_normalization_66 (Ba  (None, 12, 12, 192)          576       ['conv2d_65[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_65 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_66[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_61 (Conv2D)          (None, 12, 12, 192)          147456    ['mixed6[0][0]']              
                                                                                                  
 conv2d_66 (Conv2D)          (None, 12, 12, 192)          258048    ['activation_65[0][0]']       
                                                                                                  
 batch_normalization_62 (Ba  (None, 12, 12, 192)          576       ['conv2d_61[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_67 (Ba  (None, 12, 12, 192)          576       ['conv2d_66[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_61 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_62[0][0]
                                                                    ']                            
                                                                                                  
 activation_66 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_67[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_62 (Conv2D)          (None, 12, 12, 192)          258048    ['activation_61[0][0]']       
                                                                                                  
 conv2d_67 (Conv2D)          (None, 12, 12, 192)          258048    ['activation_66[0][0]']       
                                                                                                  
 batch_normalization_63 (Ba  (None, 12, 12, 192)          576       ['conv2d_62[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_68 (Ba  (None, 12, 12, 192)          576       ['conv2d_67[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_62 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_63[0][0]
                                                                    ']                            
                                                                                                  
 activation_67 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_68[0][0]
                                                                    ']                            
                                                                                                  
 average_pooling2d_6 (Avera  (None, 12, 12, 768)          0         ['mixed6[0][0]']              
 gePooling2D)                                                                                     
                                                                                                  
 conv2d_60 (Conv2D)          (None, 12, 12, 192)          147456    ['mixed6[0][0]']              
                                                                                                  
 conv2d_63 (Conv2D)          (None, 12, 12, 192)          258048    ['activation_62[0][0]']       
                                                                                                  
 conv2d_68 (Conv2D)          (None, 12, 12, 192)          258048    ['activation_67[0][0]']       
                                                                                                  
 conv2d_69 (Conv2D)          (None, 12, 12, 192)          147456    ['average_pooling2d_6[0][0]'] 
                                                                                                  
 batch_normalization_61 (Ba  (None, 12, 12, 192)          576       ['conv2d_60[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_64 (Ba  (None, 12, 12, 192)          576       ['conv2d_63[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_69 (Ba  (None, 12, 12, 192)          576       ['conv2d_68[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_70 (Ba  (None, 12, 12, 192)          576       ['conv2d_69[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_60 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_61[0][0]
                                                                    ']                            
                                                                                                  
 activation_63 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_64[0][0]
                                                                    ']                            
                                                                                                  
 activation_68 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_69[0][0]
                                                                    ']                            
                                                                                                  
 activation_69 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_70[0][0]
                                                                    ']                            
                                                                                                  
 mixed7 (Concatenate)        (None, 12, 12, 768)          0         ['activation_60[0][0]',       
                                                                     'activation_63[0][0]',       
                                                                     'activation_68[0][0]',       
                                                                     'activation_69[0][0]']       
                                                                                                  
 conv2d_72 (Conv2D)          (None, 12, 12, 192)          147456    ['mixed7[0][0]']              
                                                                                                  
 batch_normalization_73 (Ba  (None, 12, 12, 192)          576       ['conv2d_72[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_72 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_73[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_73 (Conv2D)          (None, 12, 12, 192)          258048    ['activation_72[0][0]']       
                                                                                                  
 batch_normalization_74 (Ba  (None, 12, 12, 192)          576       ['conv2d_73[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_73 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_74[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_70 (Conv2D)          (None, 12, 12, 192)          147456    ['mixed7[0][0]']              
                                                                                                  
 conv2d_74 (Conv2D)          (None, 12, 12, 192)          258048    ['activation_73[0][0]']       
                                                                                                  
 batch_normalization_71 (Ba  (None, 12, 12, 192)          576       ['conv2d_70[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_75 (Ba  (None, 12, 12, 192)          576       ['conv2d_74[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_70 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_71[0][0]
                                                                    ']                            
                                                                                                  
 activation_74 (Activation)  (None, 12, 12, 192)          0         ['batch_normalization_75[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_71 (Conv2D)          (None, 5, 5, 320)            552960    ['activation_70[0][0]']       
                                                                                                  
 conv2d_75 (Conv2D)          (None, 5, 5, 192)            331776    ['activation_74[0][0]']       
                                                                                                  
 batch_normalization_72 (Ba  (None, 5, 5, 320)            960       ['conv2d_71[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_76 (Ba  (None, 5, 5, 192)            576       ['conv2d_75[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_71 (Activation)  (None, 5, 5, 320)            0         ['batch_normalization_72[0][0]
                                                                    ']                            
                                                                                                  
 activation_75 (Activation)  (None, 5, 5, 192)            0         ['batch_normalization_76[0][0]
                                                                    ']                            
                                                                                                  
 max_pooling2d_3 (MaxPoolin  (None, 5, 5, 768)            0         ['mixed7[0][0]']              
 g2D)                                                                                             
                                                                                                  
 mixed8 (Concatenate)        (None, 5, 5, 1280)           0         ['activation_71[0][0]',       
                                                                     'activation_75[0][0]',       
                                                                     'max_pooling2d_3[0][0]']     
                                                                                                  
 conv2d_80 (Conv2D)          (None, 5, 5, 448)            573440    ['mixed8[0][0]']              
                                                                                                  
 batch_normalization_81 (Ba  (None, 5, 5, 448)            1344      ['conv2d_80[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_80 (Activation)  (None, 5, 5, 448)            0         ['batch_normalization_81[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_77 (Conv2D)          (None, 5, 5, 384)            491520    ['mixed8[0][0]']              
                                                                                                  
 conv2d_81 (Conv2D)          (None, 5, 5, 384)            1548288   ['activation_80[0][0]']       
                                                                                                  
 batch_normalization_78 (Ba  (None, 5, 5, 384)            1152      ['conv2d_77[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_82 (Ba  (None, 5, 5, 384)            1152      ['conv2d_81[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_77 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_78[0][0]
                                                                    ']                            
                                                                                                  
 activation_81 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_82[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_78 (Conv2D)          (None, 5, 5, 384)            442368    ['activation_77[0][0]']       
                                                                                                  
 conv2d_79 (Conv2D)          (None, 5, 5, 384)            442368    ['activation_77[0][0]']       
                                                                                                  
 conv2d_82 (Conv2D)          (None, 5, 5, 384)            442368    ['activation_81[0][0]']       
                                                                                                  
 conv2d_83 (Conv2D)          (None, 5, 5, 384)            442368    ['activation_81[0][0]']       
                                                                                                  
 average_pooling2d_7 (Avera  (None, 5, 5, 1280)           0         ['mixed8[0][0]']              
 gePooling2D)                                                                                     
                                                                                                  
 conv2d_76 (Conv2D)          (None, 5, 5, 320)            409600    ['mixed8[0][0]']              
                                                                                                  
 batch_normalization_79 (Ba  (None, 5, 5, 384)            1152      ['conv2d_78[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_80 (Ba  (None, 5, 5, 384)            1152      ['conv2d_79[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_83 (Ba  (None, 5, 5, 384)            1152      ['conv2d_82[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_84 (Ba  (None, 5, 5, 384)            1152      ['conv2d_83[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 conv2d_84 (Conv2D)          (None, 5, 5, 192)            245760    ['average_pooling2d_7[0][0]'] 
                                                                                                  
 batch_normalization_77 (Ba  (None, 5, 5, 320)            960       ['conv2d_76[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_78 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_79[0][0]
                                                                    ']                            
                                                                                                  
 activation_79 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_80[0][0]
                                                                    ']                            
                                                                                                  
 activation_82 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_83[0][0]
                                                                    ']                            
                                                                                                  
 activation_83 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_84[0][0]
                                                                    ']                            
                                                                                                  
 batch_normalization_85 (Ba  (None, 5, 5, 192)            576       ['conv2d_84[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_76 (Activation)  (None, 5, 5, 320)            0         ['batch_normalization_77[0][0]
                                                                    ']                            
                                                                                                  
 mixed9_0 (Concatenate)      (None, 5, 5, 768)            0         ['activation_78[0][0]',       
                                                                     'activation_79[0][0]']       
                                                                                                  
 concatenate (Concatenate)   (None, 5, 5, 768)            0         ['activation_82[0][0]',       
                                                                     'activation_83[0][0]']       
                                                                                                  
 activation_84 (Activation)  (None, 5, 5, 192)            0         ['batch_normalization_85[0][0]
                                                                    ']                            
                                                                                                  
 mixed9 (Concatenate)        (None, 5, 5, 2048)           0         ['activation_76[0][0]',       
                                                                     'mixed9_0[0][0]',            
                                                                     'concatenate[0][0]',         
                                                                     'activation_84[0][0]']       
                                                                                                  
 conv2d_89 (Conv2D)          (None, 5, 5, 448)            917504    ['mixed9[0][0]']              
                                                                                                  
 batch_normalization_90 (Ba  (None, 5, 5, 448)            1344      ['conv2d_89[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_89 (Activation)  (None, 5, 5, 448)            0         ['batch_normalization_90[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_86 (Conv2D)          (None, 5, 5, 384)            786432    ['mixed9[0][0]']              
                                                                                                  
 conv2d_90 (Conv2D)          (None, 5, 5, 384)            1548288   ['activation_89[0][0]']       
                                                                                                  
 batch_normalization_87 (Ba  (None, 5, 5, 384)            1152      ['conv2d_86[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_91 (Ba  (None, 5, 5, 384)            1152      ['conv2d_90[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_86 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_87[0][0]
                                                                    ']                            
                                                                                                  
 activation_90 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_91[0][0]
                                                                    ']                            
                                                                                                  
 conv2d_87 (Conv2D)          (None, 5, 5, 384)            442368    ['activation_86[0][0]']       
                                                                                                  
 conv2d_88 (Conv2D)          (None, 5, 5, 384)            442368    ['activation_86[0][0]']       
                                                                                                  
 conv2d_91 (Conv2D)          (None, 5, 5, 384)            442368    ['activation_90[0][0]']       
                                                                                                  
 conv2d_92 (Conv2D)          (None, 5, 5, 384)            442368    ['activation_90[0][0]']       
                                                                                                  
 average_pooling2d_8 (Avera  (None, 5, 5, 2048)           0         ['mixed9[0][0]']              
 gePooling2D)                                                                                     
                                                                                                  
 conv2d_85 (Conv2D)          (None, 5, 5, 320)            655360    ['mixed9[0][0]']              
                                                                                                  
 batch_normalization_88 (Ba  (None, 5, 5, 384)            1152      ['conv2d_87[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_89 (Ba  (None, 5, 5, 384)            1152      ['conv2d_88[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_92 (Ba  (None, 5, 5, 384)            1152      ['conv2d_91[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 batch_normalization_93 (Ba  (None, 5, 5, 384)            1152      ['conv2d_92[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 conv2d_93 (Conv2D)          (None, 5, 5, 192)            393216    ['average_pooling2d_8[0][0]'] 
                                                                                                  
 batch_normalization_86 (Ba  (None, 5, 5, 320)            960       ['conv2d_85[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_87 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_88[0][0]
                                                                    ']                            
                                                                                                  
 activation_88 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_89[0][0]
                                                                    ']                            
                                                                                                  
 activation_91 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_92[0][0]
                                                                    ']                            
                                                                                                  
 activation_92 (Activation)  (None, 5, 5, 384)            0         ['batch_normalization_93[0][0]
                                                                    ']                            
                                                                                                  
 batch_normalization_94 (Ba  (None, 5, 5, 192)            576       ['conv2d_93[0][0]']           
 tchNormalization)                                                                                
                                                                                                  
 activation_85 (Activation)  (None, 5, 5, 320)            0         ['batch_normalization_86[0][0]
                                                                    ']                            
                                                                                                  
 mixed9_1 (Concatenate)      (None, 5, 5, 768)            0         ['activation_87[0][0]',       
                                                                     'activation_88[0][0]']       
                                                                                                  
 concatenate_1 (Concatenate  (None, 5, 5, 768)            0         ['activation_91[0][0]',       
 )                                                                   'activation_92[0][0]']       
                                                                                                  
 activation_93 (Activation)  (None, 5, 5, 192)            0         ['batch_normalization_94[0][0]
                                                                    ']                            
                                                                                                  
 mixed10 (Concatenate)       (None, 5, 5, 2048)           0         ['activation_85[0][0]',       
                                                                     'mixed9_1[0][0]',            
                                                                     'concatenate_1[0][0]',       
                                                                     'activation_93[0][0]']       
                                                                                                  
 global_average_pooling2d_1  (None, 2048)                 0         ['mixed10[0][0]']             
  (GlobalAveragePooling2D)                                                                        
                                                                                                  
 dense_2 (Dense)             (None, 1024)                 2098176   ['global_average_pooling2d_1[0
                                                                    ][0]']                        
                                                                                                  
 dense_3 (Dense)             (None, 196)                  200900    ['dense_2[0][0]']             
                                                                                                  
==================================================================================================
Total params: 24101860 (91.94 MB)
Trainable params: 2299076 (8.77 MB)
Non-trainable params: 21802784 (83.17 MB)
__________________________________________________________________________________________________
In [52]:
# Convert to numpy arrays
# Ensure all images have the same shape before stacking
# Ensure the model summary is called after defining the model
googlenet_batch_size=16
history_googlenet= googlenet_model.fit(
    train_generator,  # Uses batches from the generator
    steps_per_epoch=len(df_train) // googlenet_batch_size,  # Number of batches per epoch
    epochs=10,
    validation_data=val_generator,  # Uses batches from the validation generator
    validation_steps=len(df_val) // googlenet_batch_size,  # Number of validation batches per epoch
)

Epoch 1/10
407/407 [==============================] - 53s 86ms/step - loss: 4.4657 - accuracy: 0.0585 - val_loss: 3.8620 - val_accuracy: 0.1017
Epoch 2/10
407/407 [==============================] - 26s 64ms/step - loss: 3.4662 - accuracy: 0.1531 - val_loss: 3.5989 - val_accuracy: 0.1426
Epoch 3/10
407/407 [==============================] - 26s 65ms/step - loss: 3.0104 - accuracy: 0.2356 - val_loss: 3.4258 - val_accuracy: 0.1841
Epoch 4/10
407/407 [==============================] - 26s 65ms/step - loss: 2.6549 - accuracy: 0.3205 - val_loss: 3.3805 - val_accuracy: 0.1866
Epoch 5/10
407/407 [==============================] - 26s 63ms/step - loss: 2.3704 - accuracy: 0.3773 - val_loss: 3.3718 - val_accuracy: 0.2015
Epoch 6/10
407/407 [==============================] - 25s 61ms/step - loss: 2.1265 - accuracy: 0.4351 - val_loss: 3.3983 - val_accuracy: 0.2126
Epoch 7/10
407/407 [==============================] - 23s 57ms/step - loss: 1.9191 - accuracy: 0.4816 - val_loss: 3.4501 - val_accuracy: 0.2139
Epoch 8/10
407/407 [==============================] - 21s 53ms/step - loss: 1.7333 - accuracy: 0.5312 - val_loss: 3.5264 - val_accuracy: 0.2219
Epoch 9/10
407/407 [==============================] - 21s 52ms/step - loss: 1.5608 - accuracy: 0.5769 - val_loss: 3.6023 - val_accuracy: 0.2343
Epoch 10/10
407/407 [==============================] - 21s 52ms/step - loss: 1.4138 - accuracy: 0.6175 - val_loss: 3.6964 - val_accuracy: 0.2393
In [53]:
#display model accuracy vs model loss
plot_training_history(history_googlenet)
In [54]:
y_pred, y_true,df_googlenet_classification_report = generate_classification_report_tf_model(
    model=googlenet_model,
    df_val=df_val,
    label_encoder=label_encoder,
    preprocess_fn=googlenet_preprocess,
    batch_size=32,
    report_name="googlenet_classification_report.csv"
)

51/51 [==============================] - 10s 43ms/step
Model Accuracy: 0.2406

Classification Report:
Report saved as: googlenet_classification_report.csv
Model Accuracy: 0.2406
Average Summary Metrics:
                  precision    recall  f1-score
macro avg          0.357037  0.240374  0.219307
weighted avg       0.384768  0.240638  0.231432
overall_accuracy   0.240638       NaN       NaN

Displaying top 10 of googlenet in confusion matrix

In [55]:
df_support = df_googlenet_classification_report.iloc[:-3]  # exclude average rows
top_10_classes = df_support.sort_values("support", ascending=False).head(10).index.tolist()
top_10_indices = [np.where(label_encoder.classes_ == cls)[0][0] for cls in top_10_classes]
googlenet_cm = confusion_matrix(y_true, y_pred)
googlenet_cm_top10 = googlenet_cm[np.ix_(top_10_indices, top_10_indices)]
In [56]:
plt.figure(figsize=(10, 8))
sns.heatmap(googlenet_cm_top10, annot=True, fmt='d',
            xticklabels=top_10_classes, yticklabels=top_10_classes,
            cmap='Blues')
plt.title("GoogleNet Confusion Matrix (Top 10 Classes)")
plt.xlabel("Predicted")
plt.ylabel("True")
plt.tight_layout()
plt.show()
Googlenet Summary¶

The inception modules allow the model to learn features at different scales, which can be beneficial for detecting cars of various sizes and orientations.

Training Accuracy: Steadily increases, reaching ~75%. Validation Accuracy: Stagnates around 20-25%, indicating poor generalization. Training Loss: Decreases smoothly, showing effective learning on training data.

Validation Loss: Plateaus and increases after a few epochs, a sign of overfitting.

Classification Report Analysis: Overall Accuracy: 25%, indicating poor performance on validation data. Precision: ~41% Recall: ~26% (Very Low) F1-Score: ~24%

Key Issues Identified: Indicates skewed performance, likely due to class imbalance. Overfitting. High Bias (Poor Performance on Validation Data) Potential Class Imbalance

6C. AlexNet

In [57]:
# Define paths
#image_dir = 'Car_Images/Car Images/Test Images'  # Adjust based on your directory structure
image_dir = 'car_data/car_data/test' 
# Prepare data
images = []
labels = []

for index, row in test_annotations_df.iterrows():
    #image_name = row['Image Name']
    image_name = row['image_name']
    # Load and preprocess the image
    image = cv2.imread(image_path)
    image = cv2.resize(image, (227, 227))  # Resize to 227x227 pixels (AlexNet input size)
    images.append(image)

    # Assuming 'Image class' contains the class label
    #labels.append(row['Image class'])
    labels.append(row['image_class'])

# Convert to numpy arrays
images = np.array(images)
labels = np.array(labels)

# Encode labels
unique_classes = np.unique(labels)
In [58]:
def create_alexnet_model(input_shape, num_classes):
    model = Sequential()

    # First Convolutional Layer
    model.add(Conv2D(96, (11, 11), strides=(4, 4), activation='relu', input_shape=input_shape))
    model.add(MaxPooling2D(pool_size=(3, 3), strides=(2, 2)))
    model.add(BatchNormalization())

    # Second Convolutional Layer
    model.add(Conv2D(256, (5, 5), padding='same', activation='relu'))
    model.add(MaxPooling2D(pool_size=(3, 3), strides=(2, 2)))
    model.add(BatchNormalization())

    # Third Convolutional Layer
    model.add(Conv2D(384, (3, 3), padding='same', activation='relu'))

    # Fourth Convolutional Layer
    model.add(Conv2D(384, (3, 3), padding='same', activation='relu'))

    # Fifth Convolutional Layer
    model.add(Conv2D(256, (3, 3), padding='same', activation='relu'))
    model.add(MaxPooling2D(pool_size=(3, 3), strides=(2, 2)))
    model.add(BatchNormalization())

    # Flatten the output
    model.add(Flatten())

    # Fully Connected Layers
    model.add(Dense(4096, activation='relu'))
    model.add(Dropout(0.5))

    model.add(Dense(4096, activation='relu'))
    model.add(Dropout(0.5))

    model.add(Dense(num_classes, activation='softmax'))

    return model

# Create the model
input_shape = (224, 224, 3)  # Image dimensions for AlexNet

#num_classes = len(unique_classes)

num_classes = len(df_training['labels'].unique())

model = create_alexnet_model(input_shape, num_classes)

model.compile(optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'])

# Data augmentation
datagen = ImageDataGenerator(rotation_range=20, width_shift_range=0.2,
                             height_shift_range=0.2, shear_range=0.2,
                             zoom_range=0.2, horizontal_flip=True,
                             fill_mode='nearest')

epochs=10
#batch_size=32
batch_size=16

train_steps = len(df_train) // batch_size
val_steps = len(df_val) // batch_size
model.summary()

alexnet_history = model.fit(
    train_generator,  
    steps_per_epoch = train_steps,
    epochs=epochs,
    batch_size=32,    
    validation_data=val_generator,
    validation_steps=val_steps
)

Model: "sequential"
_________________________________________________________________
 Layer (type)                Output Shape              Param #   
=================================================================
 conv2d_94 (Conv2D)          (None, 54, 54, 96)        34944     
                                                                 
 max_pooling2d_4 (MaxPoolin  (None, 26, 26, 96)        0         
 g2D)                                                            
                                                                 
 batch_normalization_95 (Ba  (None, 26, 26, 96)        384       
 tchNormalization)                                               
                                                                 
 conv2d_95 (Conv2D)          (None, 26, 26, 256)       614656    
                                                                 
 max_pooling2d_5 (MaxPoolin  (None, 12, 12, 256)       0         
 g2D)                                                            
                                                                 
 batch_normalization_96 (Ba  (None, 12, 12, 256)       1024      
 tchNormalization)                                               
                                                                 
 conv2d_96 (Conv2D)          (None, 12, 12, 384)       885120    
                                                                 
 conv2d_97 (Conv2D)          (None, 12, 12, 384)       1327488   
                                                                 
 conv2d_98 (Conv2D)          (None, 12, 12, 256)       884992    
                                                                 
 max_pooling2d_6 (MaxPoolin  (None, 5, 5, 256)         0         
 g2D)                                                            
                                                                 
 batch_normalization_97 (Ba  (None, 5, 5, 256)         1024      
 tchNormalization)                                               
                                                                 
 flatten (Flatten)           (None, 6400)              0         
                                                                 
 dense_4 (Dense)             (None, 4096)              26218496  
                                                                 
 dropout_1 (Dropout)         (None, 4096)              0         
                                                                 
 dense_5 (Dense)             (None, 4096)              16781312  
                                                                 
 dropout_2 (Dropout)         (None, 4096)              0         
                                                                 
 dense_6 (Dense)             (None, 196)               803012    
                                                                 
=================================================================
Total params: 47552452 (181.40 MB)
Trainable params: 47551236 (181.39 MB)
Non-trainable params: 1216 (4.75 KB)
_________________________________________________________________
Epoch 1/10
407/407 [==============================] - 32s 67ms/step - loss: 5.5415 - accuracy: 0.0051 - val_loss: 5.2992 - val_accuracy: 0.0087
Epoch 2/10
407/407 [==============================] - 26s 64ms/step - loss: 5.2811 - accuracy: 0.0068 - val_loss: 5.2869 - val_accuracy: 0.0087
Epoch 3/10
407/407 [==============================] - 26s 63ms/step - loss: 5.2773 - accuracy: 0.0082 - val_loss: 5.2899 - val_accuracy: 0.0087
Epoch 4/10
407/407 [==============================] - 25s 62ms/step - loss: 5.2803 - accuracy: 0.0083 - val_loss: 5.2937 - val_accuracy: 0.0087
Epoch 5/10
407/407 [==============================] - 25s 60ms/step - loss: 5.2759 - accuracy: 0.0083 - val_loss: 5.2935 - val_accuracy: 0.0087
Epoch 6/10
407/407 [==============================] - 23s 57ms/step - loss: 5.2741 - accuracy: 0.0083 - val_loss: 5.2959 - val_accuracy: 0.0087
Epoch 7/10
407/407 [==============================] - 21s 52ms/step - loss: 5.2738 - accuracy: 0.0083 - val_loss: 5.2961 - val_accuracy: 0.0087
Epoch 8/10
407/407 [==============================] - 21s 51ms/step - loss: 5.2734 - accuracy: 0.0083 - val_loss: 5.2964 - val_accuracy: 0.0087
Epoch 9/10
407/407 [==============================] - 21s 52ms/step - loss: 5.2731 - accuracy: 0.0083 - val_loss: 5.2975 - val_accuracy: 0.0087
Epoch 10/10
407/407 [==============================] - 21s 52ms/step - loss: 5.2725 - accuracy: 0.0083 - val_loss: 5.2986 - val_accuracy: 0.0087
In [59]:
#display model accuracy vs loss
plot_training_history(alexnet_history)
In [60]:
X_val = np.array([img for img in df_val['image']])
y_val_true = np.array([np.argmax(label) for label in df_val['label_categorical']])

# Predict in one go
y_val_pred = np.argmax(model.predict(X_val), axis=1)

51/51 [==============================] - 1s 11ms/step
In [61]:
alexnet_report = classification_report(
    y_val_true,
    y_val_pred,
    target_names=label_encoder.classes_,
    output_dict=True,
    zero_division=1
)
acc = accuracy_score(y_val_true, y_val_pred)
In [62]:
df_alexnet_classification_report = pd.DataFrame(alexnet_report).transpose()
df_alexnet_classification_report.loc["overall_accuracy"] = [acc, None, None, None]
In [63]:
df_alexnet_classification_report.to_csv("alexnet_classification_report_vectorized.csv")
In [64]:
print(f"Accuracy Score: {acc:.4f}")
print("Average Summary Metrics:")
print(df_alexnet_classification_report.tail(3)[["precision", "recall", "f1-score"]])

Accuracy Score: 0.0086
Average Summary Metrics:
                  precision    recall  f1-score
macro avg          0.989840  0.005102  0.000087
weighted avg       0.987796  0.008594  0.000147
overall_accuracy   0.008594       NaN       NaN
In [65]:
# Print Classification Report
#print("Classification Report:")
#print(classification_report(y_val_true, y_val_pred, target_names=df_training['labels'].unique(), zero_division=0))

confusion metrics

In [66]:
df_support = df_alexnet_classification_report.iloc[:-3]  # exclude average rows
top_10_classes = df_support.sort_values("support", ascending=False).head(10).index.tolist()
top_10_indices = [np.where(label_encoder.classes_ == cls)[0][0] for cls in top_10_classes]
alexnet_cm = confusion_matrix(y_val_true, y_val_pred)
alexnet_cm_top10 = alexnet_cm[np.ix_(top_10_indices, top_10_indices)]
In [67]:
plt.figure(figsize=(10, 8))
sns.heatmap(alexnet_cm_top10, annot=True, fmt='d',
            xticklabels=top_10_classes, yticklabels=top_10_classes,
            cmap='Blues')
plt.title("AlexNet Confusion Matrix (Top 10 Classes)")
plt.xlabel("Predicted")
plt.ylabel("True")
plt.tight_layout()
plt.show()

Observation For Alexnet Model¶

  • The Model is not learning appropriately since the Training Accuracy and Validation accuracy are pretty low
  • There is huge value loss in both training as well as validation datasets which indicates that the model is not performing well
  • We have the accuracy reported as none in the classification report. which inidicates that the model is not performing well

Further Actions that can be taken are

  • Increaase Data Augmentation for the model to perform well, in the training data set.
  • Possible class imbalance
  • Parameter tuning

6D. ResNet

In [68]:
# Load ResNet50 base model without the top layer
base_model = ResNet50(weights='imagenet', include_top=False, input_shape=(224, 224, 3))

# Freeze base model layers
for layer in base_model.layers:
    layer.trainable = False

# Add custom classification layers
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(512, activation='relu')(x)
x = Dropout(0.5)(x)
x = Dense(len(df_training['labels_encoded'].unique()), activation='softmax')(x)  # Output layer

# Define model
resnet_model = Model(inputs=base_model.input, outputs=x)

# Compile model
resnet_model.compile(optimizer=Adam(learning_rate=0.0001), loss='categorical_crossentropy', metrics=['accuracy'])

# Print model summary
resnet_model.summary()

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/resnet/resnet50_weights_tf_dim_ordering_tf_kernels_notop.h5
94765736/94765736 [==============================] - 7s 0us/step
Model: "model_2"
__________________________________________________________________________________________________
 Layer (type)                Output Shape                 Param #   Connected to                  
==================================================================================================
 input_3 (InputLayer)        [(None, 224, 224, 3)]        0         []                            
                                                                                                  
 conv1_pad (ZeroPadding2D)   (None, 230, 230, 3)          0         ['input_3[0][0]']             
                                                                                                  
 conv1_conv (Conv2D)         (None, 112, 112, 64)         9472      ['conv1_pad[0][0]']           
                                                                                                  
 conv1_bn (BatchNormalizati  (None, 112, 112, 64)         256       ['conv1_conv[0][0]']          
 on)                                                                                              
                                                                                                  
 conv1_relu (Activation)     (None, 112, 112, 64)         0         ['conv1_bn[0][0]']            
                                                                                                  
 pool1_pad (ZeroPadding2D)   (None, 114, 114, 64)         0         ['conv1_relu[0][0]']          
                                                                                                  
 pool1_pool (MaxPooling2D)   (None, 56, 56, 64)           0         ['pool1_pad[0][0]']           
                                                                                                  
 conv2_block1_1_conv (Conv2  (None, 56, 56, 64)           4160      ['pool1_pool[0][0]']          
 D)                                                                                               
                                                                                                  
 conv2_block1_1_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block1_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block1_1_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block1_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block1_2_conv (Conv2  (None, 56, 56, 64)           36928     ['conv2_block1_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block1_2_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block1_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block1_2_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block1_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block1_0_conv (Conv2  (None, 56, 56, 256)          16640     ['pool1_pool[0][0]']          
 D)                                                                                               
                                                                                                  
 conv2_block1_3_conv (Conv2  (None, 56, 56, 256)          16640     ['conv2_block1_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block1_0_bn (BatchNo  (None, 56, 56, 256)          1024      ['conv2_block1_0_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block1_3_bn (BatchNo  (None, 56, 56, 256)          1024      ['conv2_block1_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block1_add (Add)      (None, 56, 56, 256)          0         ['conv2_block1_0_bn[0][0]',   
                                                                     'conv2_block1_3_bn[0][0]']   
                                                                                                  
 conv2_block1_out (Activati  (None, 56, 56, 256)          0         ['conv2_block1_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv2_block2_1_conv (Conv2  (None, 56, 56, 64)           16448     ['conv2_block1_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv2_block2_1_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block2_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block2_1_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block2_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block2_2_conv (Conv2  (None, 56, 56, 64)           36928     ['conv2_block2_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block2_2_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block2_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block2_2_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block2_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block2_3_conv (Conv2  (None, 56, 56, 256)          16640     ['conv2_block2_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block2_3_bn (BatchNo  (None, 56, 56, 256)          1024      ['conv2_block2_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block2_add (Add)      (None, 56, 56, 256)          0         ['conv2_block1_out[0][0]',    
                                                                     'conv2_block2_3_bn[0][0]']   
                                                                                                  
 conv2_block2_out (Activati  (None, 56, 56, 256)          0         ['conv2_block2_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv2_block3_1_conv (Conv2  (None, 56, 56, 64)           16448     ['conv2_block2_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv2_block3_1_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block3_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block3_1_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block3_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block3_2_conv (Conv2  (None, 56, 56, 64)           36928     ['conv2_block3_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block3_2_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block3_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block3_2_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block3_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block3_3_conv (Conv2  (None, 56, 56, 256)          16640     ['conv2_block3_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block3_3_bn (BatchNo  (None, 56, 56, 256)          1024      ['conv2_block3_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block3_add (Add)      (None, 56, 56, 256)          0         ['conv2_block2_out[0][0]',    
                                                                     'conv2_block3_3_bn[0][0]']   
                                                                                                  
 conv2_block3_out (Activati  (None, 56, 56, 256)          0         ['conv2_block3_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv3_block1_1_conv (Conv2  (None, 28, 28, 128)          32896     ['conv2_block3_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv3_block1_1_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block1_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block1_1_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block1_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block1_2_conv (Conv2  (None, 28, 28, 128)          147584    ['conv3_block1_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block1_2_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block1_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block1_2_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block1_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block1_0_conv (Conv2  (None, 28, 28, 512)          131584    ['conv2_block3_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv3_block1_3_conv (Conv2  (None, 28, 28, 512)          66048     ['conv3_block1_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block1_0_bn (BatchNo  (None, 28, 28, 512)          2048      ['conv3_block1_0_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block1_3_bn (BatchNo  (None, 28, 28, 512)          2048      ['conv3_block1_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block1_add (Add)      (None, 28, 28, 512)          0         ['conv3_block1_0_bn[0][0]',   
                                                                     'conv3_block1_3_bn[0][0]']   
                                                                                                  
 conv3_block1_out (Activati  (None, 28, 28, 512)          0         ['conv3_block1_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv3_block2_1_conv (Conv2  (None, 28, 28, 128)          65664     ['conv3_block1_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv3_block2_1_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block2_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block2_1_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block2_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block2_2_conv (Conv2  (None, 28, 28, 128)          147584    ['conv3_block2_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block2_2_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block2_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block2_2_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block2_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block2_3_conv (Conv2  (None, 28, 28, 512)          66048     ['conv3_block2_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block2_3_bn (BatchNo  (None, 28, 28, 512)          2048      ['conv3_block2_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block2_add (Add)      (None, 28, 28, 512)          0         ['conv3_block1_out[0][0]',    
                                                                     'conv3_block2_3_bn[0][0]']   
                                                                                                  
 conv3_block2_out (Activati  (None, 28, 28, 512)          0         ['conv3_block2_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv3_block3_1_conv (Conv2  (None, 28, 28, 128)          65664     ['conv3_block2_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv3_block3_1_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block3_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block3_1_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block3_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block3_2_conv (Conv2  (None, 28, 28, 128)          147584    ['conv3_block3_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block3_2_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block3_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block3_2_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block3_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block3_3_conv (Conv2  (None, 28, 28, 512)          66048     ['conv3_block3_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block3_3_bn (BatchNo  (None, 28, 28, 512)          2048      ['conv3_block3_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block3_add (Add)      (None, 28, 28, 512)          0         ['conv3_block2_out[0][0]',    
                                                                     'conv3_block3_3_bn[0][0]']   
                                                                                                  
 conv3_block3_out (Activati  (None, 28, 28, 512)          0         ['conv3_block3_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv3_block4_1_conv (Conv2  (None, 28, 28, 128)          65664     ['conv3_block3_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv3_block4_1_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block4_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block4_1_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block4_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block4_2_conv (Conv2  (None, 28, 28, 128)          147584    ['conv3_block4_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block4_2_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block4_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block4_2_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block4_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block4_3_conv (Conv2  (None, 28, 28, 512)          66048     ['conv3_block4_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block4_3_bn (BatchNo  (None, 28, 28, 512)          2048      ['conv3_block4_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block4_add (Add)      (None, 28, 28, 512)          0         ['conv3_block3_out[0][0]',    
                                                                     'conv3_block4_3_bn[0][0]']   
                                                                                                  
 conv3_block4_out (Activati  (None, 28, 28, 512)          0         ['conv3_block4_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block1_1_conv (Conv2  (None, 14, 14, 256)          131328    ['conv3_block4_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block1_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block1_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block1_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block1_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block1_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block1_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block1_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block1_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block1_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block1_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block1_0_conv (Conv2  (None, 14, 14, 1024)         525312    ['conv3_block4_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block1_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block1_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block1_0_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block1_0_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block1_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block1_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block1_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block1_0_bn[0][0]',   
                                                                     'conv4_block1_3_bn[0][0]']   
                                                                                                  
 conv4_block1_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block1_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block2_1_conv (Conv2  (None, 14, 14, 256)          262400    ['conv4_block1_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block2_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block2_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block2_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block2_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block2_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block2_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block2_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block2_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block2_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block2_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block2_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block2_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block2_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block2_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block2_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block1_out[0][0]',    
                                                                     'conv4_block2_3_bn[0][0]']   
                                                                                                  
 conv4_block2_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block2_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block3_1_conv (Conv2  (None, 14, 14, 256)          262400    ['conv4_block2_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block3_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block3_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block3_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block3_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block3_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block3_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block3_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block3_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block3_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block3_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block3_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block3_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block3_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block3_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block3_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block2_out[0][0]',    
                                                                     'conv4_block3_3_bn[0][0]']   
                                                                                                  
 conv4_block3_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block3_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block4_1_conv (Conv2  (None, 14, 14, 256)          262400    ['conv4_block3_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block4_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block4_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block4_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block4_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block4_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block4_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block4_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block4_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block4_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block4_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block4_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block4_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block4_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block4_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block4_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block3_out[0][0]',    
                                                                     'conv4_block4_3_bn[0][0]']   
                                                                                                  
 conv4_block4_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block4_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block5_1_conv (Conv2  (None, 14, 14, 256)          262400    ['conv4_block4_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block5_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block5_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block5_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block5_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block5_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block5_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block5_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block5_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block5_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block5_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block5_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block5_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block5_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block5_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block5_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block4_out[0][0]',    
                                                                     'conv4_block5_3_bn[0][0]']   
                                                                                                  
 conv4_block5_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block5_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block6_1_conv (Conv2  (None, 14, 14, 256)          262400    ['conv4_block5_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block6_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block6_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block6_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block6_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block6_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block6_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block6_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block6_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block6_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block6_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block6_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block6_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block6_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block6_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block6_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block5_out[0][0]',    
                                                                     'conv4_block6_3_bn[0][0]']   
                                                                                                  
 conv4_block6_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block6_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv5_block1_1_conv (Conv2  (None, 7, 7, 512)            524800    ['conv4_block6_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv5_block1_1_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block1_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block1_1_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block1_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block1_2_conv (Conv2  (None, 7, 7, 512)            2359808   ['conv5_block1_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block1_2_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block1_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block1_2_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block1_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block1_0_conv (Conv2  (None, 7, 7, 2048)           2099200   ['conv4_block6_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv5_block1_3_conv (Conv2  (None, 7, 7, 2048)           1050624   ['conv5_block1_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block1_0_bn (BatchNo  (None, 7, 7, 2048)           8192      ['conv5_block1_0_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block1_3_bn (BatchNo  (None, 7, 7, 2048)           8192      ['conv5_block1_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block1_add (Add)      (None, 7, 7, 2048)           0         ['conv5_block1_0_bn[0][0]',   
                                                                     'conv5_block1_3_bn[0][0]']   
                                                                                                  
 conv5_block1_out (Activati  (None, 7, 7, 2048)           0         ['conv5_block1_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv5_block2_1_conv (Conv2  (None, 7, 7, 512)            1049088   ['conv5_block1_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv5_block2_1_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block2_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block2_1_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block2_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block2_2_conv (Conv2  (None, 7, 7, 512)            2359808   ['conv5_block2_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block2_2_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block2_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block2_2_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block2_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block2_3_conv (Conv2  (None, 7, 7, 2048)           1050624   ['conv5_block2_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block2_3_bn (BatchNo  (None, 7, 7, 2048)           8192      ['conv5_block2_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block2_add (Add)      (None, 7, 7, 2048)           0         ['conv5_block1_out[0][0]',    
                                                                     'conv5_block2_3_bn[0][0]']   
                                                                                                  
 conv5_block2_out (Activati  (None, 7, 7, 2048)           0         ['conv5_block2_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv5_block3_1_conv (Conv2  (None, 7, 7, 512)            1049088   ['conv5_block2_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv5_block3_1_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block3_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block3_1_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block3_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block3_2_conv (Conv2  (None, 7, 7, 512)            2359808   ['conv5_block3_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block3_2_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block3_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block3_2_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block3_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block3_3_conv (Conv2  (None, 7, 7, 2048)           1050624   ['conv5_block3_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block3_3_bn (BatchNo  (None, 7, 7, 2048)           8192      ['conv5_block3_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block3_add (Add)      (None, 7, 7, 2048)           0         ['conv5_block2_out[0][0]',    
                                                                     'conv5_block3_3_bn[0][0]']   
                                                                                                  
 conv5_block3_out (Activati  (None, 7, 7, 2048)           0         ['conv5_block3_add[0][0]']    
 on)                                                                                              
                                                                                                  
 global_average_pooling2d_2  (None, 2048)                 0         ['conv5_block3_out[0][0]']    
  (GlobalAveragePooling2D)                                                                        
                                                                                                  
 dense_7 (Dense)             (None, 512)                  1049088   ['global_average_pooling2d_2[0
                                                                    ][0]']                        
                                                                                                  
 dropout_3 (Dropout)         (None, 512)                  0         ['dense_7[0][0]']             
                                                                                                  
 dense_8 (Dense)             (None, 196)                  100548    ['dropout_3[0][0]']           
                                                                                                  
==================================================================================================
Total params: 24737348 (94.37 MB)
Trainable params: 1149636 (4.39 MB)
Non-trainable params: 23587712 (89.98 MB)
__________________________________________________________________________________________________
In [69]:
epochs = 10
batch_size=16
steps_per_epoch = len(df_train) // batch_size
validation_steps = len(df_val) // batch_size

resnet_history = resnet_model.fit(
    train_generator,
    steps_per_epoch=steps_per_epoch,
    validation_data=val_generator,
    validation_steps=validation_steps,
    epochs=epochs
)

Epoch 1/10
407/407 [==============================] - 40s 75ms/step - loss: 5.3837 - accuracy: 0.0068 - val_loss: 5.2926 - val_accuracy: 0.0087
Epoch 2/10
407/407 [==============================] - 26s 64ms/step - loss: 5.2883 - accuracy: 0.0060 - val_loss: 5.2853 - val_accuracy: 0.0074
Epoch 3/10
407/407 [==============================] - 26s 64ms/step - loss: 5.2770 - accuracy: 0.0078 - val_loss: 5.2850 - val_accuracy: 0.0099
Epoch 4/10
407/407 [==============================] - 25s 62ms/step - loss: 5.2739 - accuracy: 0.0075 - val_loss: 5.2816 - val_accuracy: 0.0019
Epoch 5/10
407/407 [==============================] - 25s 60ms/step - loss: 5.2711 - accuracy: 0.0069 - val_loss: 5.2787 - val_accuracy: 0.0025
Epoch 6/10
407/407 [==============================] - 23s 57ms/step - loss: 5.2674 - accuracy: 0.0075 - val_loss: 5.2804 - val_accuracy: 0.0025
Epoch 7/10
407/407 [==============================] - 21s 52ms/step - loss: 5.2624 - accuracy: 0.0100 - val_loss: 5.2796 - val_accuracy: 0.0050
Epoch 8/10
407/407 [==============================] - 21s 52ms/step - loss: 5.2574 - accuracy: 0.0091 - val_loss: 5.2800 - val_accuracy: 0.0068
Epoch 9/10
407/407 [==============================] - 21s 52ms/step - loss: 5.2519 - accuracy: 0.0097 - val_loss: 5.2781 - val_accuracy: 0.0087
Epoch 10/10
407/407 [==============================] - 21s 52ms/step - loss: 5.2459 - accuracy: 0.0089 - val_loss: 5.2707 - val_accuracy: 0.0105
In [70]:
#accuracy loss graph
plot_training_history(resnet_history)
In [71]:
y_pred, y_true,df_resnet_classification_report = generate_classification_report_tf_model(
    model=resnet_model,
    df_val=df_val,
    label_encoder=label_encoder,
    preprocess_fn=resnet_preprocess,
    batch_size=32,
    report_name="resnet_classification_report.csv"
)

51/51 [==============================] - 7s 44ms/step
Model Accuracy: 0.0104

Classification Report:
Report saved as: resnet_classification_report.csv
Model Accuracy: 0.0104
Average Summary Metrics:
                  precision    recall  f1-score
macro avg          0.989922  0.007070  0.000247
weighted avg       0.985440  0.010436  0.000340
overall_accuracy   0.010436       NaN       NaN
In [72]:
# Compute confusion matrix
df_support = df_resnet_classification_report.iloc[:-3]  # exclude average rows
top_10_classes = df_support.sort_values("support", ascending=False).head(10).index.tolist()
top_10_indices = [np.where(label_encoder.classes_ == cls)[0][0] for cls in top_10_classes]
resnet_cm = confusion_matrix(y_val_true, y_val_pred)
resnet_cm_top10 = resnet_cm[np.ix_(top_10_indices, top_10_indices)]
# Plot confusion matrix
plt.figure(figsize=(10, 8))
sns.heatmap(resnet_cm_top10, annot=True, fmt='d',
            xticklabels=top_10_classes, yticklabels=top_10_classes,
            cmap='Blues')
#sns.heatmap(cm, annot=True, fmt="d", cmap="Blues", xticklabels=label_encoder.classes_, yticklabels=label_encoder.classes_)
plt.title("ResNet Confusion Matrix (Top 10 Classes)")
plt.xlabel("Predicted Label")
plt.ylabel("True Label")
plt.show()

Observation:

The model is not doing well due to

  • exteremely low accuracy
  • High Precision and Very low Recall

Further Actions could be

  • Check for Data Imbalance
  • Fine tune parameters and retrain the model

7. Intermediate Summary - Further Steps¶

Googlenet and Resnet further in next milestone will undergo hyper parameter tunning as commonly data imbalance and accuracy is less compared to loss

Mobilenet and Alexnet are light weight models/Shallow models, hence they are being dropped from further fine tuning and comparing them with other models.

8. Fine Tuning¶

Fine Tuning Of GoogleNet Model¶

In [73]:
from tensorflow.keras.mixed_precision import set_global_policy
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint,ReduceLROnPlateau
import random
set_global_policy('mixed_float16')

INFO:tensorflow:Mixed precision compatibility check (mixed_float16): OK
Your GPU will likely run quickly with dtype policy mixed_float16 as it has compute capability of at least 7.0. Your GPU: NVIDIA A10G, compute capability 8.6
In [74]:
print("TF Version:", tf.__version__)
print("GPU Available:", tf.config.list_physical_devices('GPU'))

TF Version: 2.16.2
GPU Available: [PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]
In [75]:
unique_classes = df_training['labels'].unique()
In [76]:
base_model = InceptionV3(
    weights='imagenet',
    include_top=False,
    input_shape=(224, 224, 3)
)
In [77]:
for layer in base_model.layers[:-50]:
    layer.trainable = False
In [78]:
x = base_model.output
x = layers.GlobalAveragePooling2D()(x)
x = layers.Dense(512, activation='relu')(x)
x = tf.keras.layers.Dropout(0.5)(x)
output = layers.Dense(196, activation='softmax', dtype='float32')(x)  # Force output to float32
In [79]:
googlenet_model_tuned = Model(inputs=base_model.input, outputs=output)
In [80]:
googlenet_model_tuned.compile(
    optimizer=Adam(learning_rate=1e-3),
    loss='sparse_categorical_crossentropy',  # use categorical_crossentropy if labels are one-hot
    metrics=['accuracy', tf.keras.metrics.TopKCategoricalAccuracy(k=5)]
)
In [81]:
batch_size = 16
use_augmentation = True
In [82]:
df_split = df_training.drop(columns=['image']).copy()
In [83]:
df_train_googlenet, df_val_googlenet = train_test_split(  df_split, test_size=0.2, random_state=42)
In [84]:
train_paths = df_train_googlenet["Image_Path"].values
train_labels = np.array([np.argmax(label) for label in df_train_googlenet["label_categorical"]])
In [85]:
val_paths = df_val_googlenet["Image_Path"].values
val_labels = np.array([np.argmax(label) for label in df_val_googlenet["label_categorical"]])
In [86]:
data_augmentation = tf.keras.Sequential([
    layers.Rescaling(1./255),
    layers.RandomFlip("horizontal"),
    layers.RandomRotation(0.1),
    layers.RandomZoom(0.1),
    layers.RandomContrast(0.1),
    layers.RandomTranslation(0.1, 0.1)
])
In [87]:
def load_and_preprocess(path, label):
    image = tf.io.read_file(path)
    image = tf.image.decode_jpeg(image, channels=3)
    image = tf.image.resize(image, [224, 224])
    image = tf.cast(image, tf.float32) / 255.0
    return image, label
In [88]:
def load_preprocess_with_augment(path, label):
    image, label = load_and_preprocess(path, label)
    image = data_augmentation(image)
    return image, label
In [89]:
# Training dataset
train_ds = tf.data.Dataset.from_tensor_slices((train_paths, train_labels)).shuffle(1000)
In [90]:
# Apply map function based on flag
if use_augmentation:
    train_ds = train_ds.map(load_preprocess_with_augment, num_parallel_calls=tf.data.AUTOTUNE)
else:
    train_ds = train_ds.map(load_and_preprocess, num_parallel_calls=tf.data.AUTOTUNE)
In [91]:
train_ds = train_ds.batch(batch_size).prefetch(tf.data.AUTOTUNE)
In [92]:
val_ds = tf.data.Dataset.from_tensor_slices((val_paths, val_labels)) \
    .map(load_and_preprocess, num_parallel_calls=tf.data.AUTOTUNE) \
    .batch(batch_size) \
    .prefetch(tf.data.AUTOTUNE)
In [93]:
callbacks = [
    EarlyStopping(
        monitor='val_loss',
        patience=30,
        restore_best_weights=True,
        verbose=1
    ),
    ModelCheckpoint(
        filepath='googlenet_finetuned_best.keras',
        monitor='val_loss',
        save_best_only=True,
        verbose=1
    ),
     ReduceLROnPlateau(monitor='val_loss', factor=0.5, patience=5, min_lr=1e-7, verbose=1)
]
In [94]:
train_class_indices = np.array([np.argmax(label) for label in df_train_googlenet["label_categorical"]]) #Get Class Indicies
In [95]:
# Compute class weights
class_weights_array = compute_class_weight(
    class_weight='balanced',
    classes=np.unique(train_class_indices),
    y=train_class_indices
)

class_weights = dict(enumerate(class_weights_array)) #converting to dict
In [96]:
history = googlenet_model_tuned.fit(
    train_ds,
    validation_data=val_ds,
    epochs=20
    ,callbacks=callbacks
    ,class_weight=class_weights  #class imbalance
)

Epoch 1/20
WARNING:tensorflow:AutoGraph could not transform <function create_autocast_variable at 0x7f7889cc9bd0> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output.
Cause: <gast.gast.Expr object at 0x7f7739b9f8e0>
To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
WARNING: AutoGraph could not transform <function create_autocast_variable at 0x7f7889cc9bd0> and will run it as-is.
Please report this to the TensorFlow team. When filing the bug, set the verbosity to 10 (on Linux, `export AUTOGRAPH_VERBOSITY=10`) and attach the full output.
Cause: <gast.gast.Expr object at 0x7f7739b9f8e0>
To silence this warning, decorate the function with @tf.autograph.experimental.do_not_convert
408/408 [==============================] - ETA: 0s - loss: 5.2980 - accuracy: 0.0032 - top_k_categorical_accuracy: 0.0048             
Epoch 1: val_loss improved from inf to 5.27864, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 59s 59ms/step - loss: 5.2980 - accuracy: 0.0032 - top_k_categorical_accuracy: 0.0048 - val_loss: 5.2786 - val_accuracy: 0.0043 - val_top_k_categorical_accuracy: 0.0074 - lr: 0.0010
Epoch 2/20
408/408 [==============================] - ETA: 0s - loss: 5.2797 - accuracy: 0.0032 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 2: val_loss improved from 5.27864 to 5.27856, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 14s 35ms/step - loss: 5.2797 - accuracy: 0.0032 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2786 - val_accuracy: 0.0055 - val_top_k_categorical_accuracy: 0.0080 - lr: 0.0010
Epoch 3/20
408/408 [==============================] - ETA: 0s - loss: 5.2798 - accuracy: 0.0029 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 3: val_loss improved from 5.27856 to 5.27852, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 15s 36ms/step - loss: 5.2798 - accuracy: 0.0029 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2785 - val_accuracy: 0.0037 - val_top_k_categorical_accuracy: 0.0074 - lr: 0.0010
Epoch 4/20
408/408 [==============================] - ETA: 0s - loss: 5.2798 - accuracy: 0.0045 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 4: val_loss improved from 5.27852 to 5.27835, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 15s 36ms/step - loss: 5.2798 - accuracy: 0.0045 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2784 - val_accuracy: 0.0037 - val_top_k_categorical_accuracy: 0.0061 - lr: 0.0010
Epoch 5/20
405/408 [============================>.] - ETA: 0s - loss: 5.2805 - accuracy: 0.0029 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 5: val_loss improved from 5.27835 to 5.27826, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 14s 35ms/step - loss: 5.2798 - accuracy: 0.0029 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2783 - val_accuracy: 0.0043 - val_top_k_categorical_accuracy: 0.0061 - lr: 0.0010
Epoch 6/20
408/408 [==============================] - ETA: 0s - loss: 5.2798 - accuracy: 0.0026 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 6: val_loss did not improve from 5.27826
408/408 [==============================] - 13s 32ms/step - loss: 5.2798 - accuracy: 0.0026 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2783 - val_accuracy: 0.0043 - val_top_k_categorical_accuracy: 0.0061 - lr: 0.0010
Epoch 7/20
408/408 [==============================] - ETA: 0s - loss: 5.2797 - accuracy: 0.0035 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 7: val_loss improved from 5.27826 to 5.27826, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 14s 35ms/step - loss: 5.2797 - accuracy: 0.0035 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2783 - val_accuracy: 0.0037 - val_top_k_categorical_accuracy: 0.0049 - lr: 0.0010
Epoch 8/20
406/408 [============================>.] - ETA: 0s - loss: 5.2795 - accuracy: 0.0034 - top_k_categorical_accuracy: 0.2660         
Epoch 8: val_loss improved from 5.27826 to 5.27818, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 14s 35ms/step - loss: 5.2798 - accuracy: 0.0034 - top_k_categorical_accuracy: 0.2682 - val_loss: 5.2782 - val_accuracy: 0.0037 - val_top_k_categorical_accuracy: 0.0123 - lr: 0.0010
Epoch 9/20
408/408 [==============================] - ETA: 0s - loss: 5.2798 - accuracy: 0.0035 - top_k_categorical_accuracy: 0.4744     
Epoch 9: val_loss did not improve from 5.27818
408/408 [==============================] - 13s 32ms/step - loss: 5.2798 - accuracy: 0.0035 - top_k_categorical_accuracy: 0.4744 - val_loss: 5.2783 - val_accuracy: 0.0043 - val_top_k_categorical_accuracy: 0.0123 - lr: 0.0010
Epoch 10/20
406/408 [============================>.] - ETA: 0s - loss: 5.2788 - accuracy: 0.0025 - top_k_categorical_accuracy: 0.0468     
Epoch 10: val_loss improved from 5.27818 to 5.27813, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 15s 36ms/step - loss: 5.2798 - accuracy: 0.0025 - top_k_categorical_accuracy: 0.0467 - val_loss: 5.2781 - val_accuracy: 0.0031 - val_top_k_categorical_accuracy: 0.0104 - lr: 0.0010
Epoch 11/20
406/408 [============================>.] - ETA: 0s - loss: 5.2796 - accuracy: 0.0025 - top_k_categorical_accuracy: 0.2094         
Epoch 11: val_loss improved from 5.27813 to 5.27806, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 14s 35ms/step - loss: 5.2798 - accuracy: 0.0025 - top_k_categorical_accuracy: 0.2087 - val_loss: 5.2781 - val_accuracy: 0.0031 - val_top_k_categorical_accuracy: 0.0092 - lr: 0.0010
Epoch 12/20
408/408 [==============================] - ETA: 0s - loss: 5.2797 - accuracy: 0.0040 - top_k_categorical_accuracy: 0.3659         
Epoch 12: val_loss did not improve from 5.27806
408/408 [==============================] - 13s 32ms/step - loss: 5.2797 - accuracy: 0.0040 - top_k_categorical_accuracy: 0.3659 - val_loss: 5.2781 - val_accuracy: 0.0037 - val_top_k_categorical_accuracy: 0.0098 - lr: 0.0010
Epoch 13/20
406/408 [============================>.] - ETA: 0s - loss: 5.2792 - accuracy: 0.0028 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 13: val_loss improved from 5.27806 to 5.27805, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 14s 35ms/step - loss: 5.2798 - accuracy: 0.0028 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2781 - val_accuracy: 0.0031 - val_top_k_categorical_accuracy: 0.0086 - lr: 0.0010
Epoch 14/20
407/408 [============================>.] - ETA: 0s - loss: 5.2800 - accuracy: 0.0048 - top_k_categorical_accuracy: 0.1032         
Epoch 14: val_loss did not improve from 5.27805
408/408 [==============================] - 13s 32ms/step - loss: 5.2798 - accuracy: 0.0048 - top_k_categorical_accuracy: 0.1031 - val_loss: 5.2781 - val_accuracy: 0.0031 - val_top_k_categorical_accuracy: 0.0086 - lr: 0.0010
Epoch 15/20
408/408 [==============================] - ETA: 0s - loss: 5.2798 - accuracy: 0.0028 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 15: val_loss improved from 5.27805 to 5.27799, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 15s 36ms/step - loss: 5.2798 - accuracy: 0.0028 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2780 - val_accuracy: 0.0037 - val_top_k_categorical_accuracy: 0.0074 - lr: 0.0010
Epoch 16/20
406/408 [============================>.] - ETA: 0s - loss: 5.2789 - accuracy: 0.0037 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 16: val_loss did not improve from 5.27799

Epoch 16: ReduceLROnPlateau reducing learning rate to 0.0005000000237487257.
408/408 [==============================] - 13s 32ms/step - loss: 5.2797 - accuracy: 0.0037 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2780 - val_accuracy: 0.0043 - val_top_k_categorical_accuracy: 0.0068 - lr: 0.0010
Epoch 17/20
406/408 [============================>.] - ETA: 0s - loss: 5.2794 - accuracy: 0.0029 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 17: val_loss did not improve from 5.27799
408/408 [==============================] - 13s 33ms/step - loss: 5.2790 - accuracy: 0.0029 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2780 - val_accuracy: 0.0043 - val_top_k_categorical_accuracy: 0.0055 - lr: 5.0000e-04
Epoch 18/20
407/408 [============================>.] - ETA: 0s - loss: 5.2788 - accuracy: 0.0043 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 18: val_loss did not improve from 5.27799
408/408 [==============================] - 13s 32ms/step - loss: 5.2790 - accuracy: 0.0043 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2780 - val_accuracy: 0.0043 - val_top_k_categorical_accuracy: 0.0061 - lr: 5.0000e-04
Epoch 19/20
408/408 [==============================] - ETA: 0s - loss: 5.2790 - accuracy: 0.0043 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 19: val_loss did not improve from 5.27799
408/408 [==============================] - 13s 32ms/step - loss: 5.2790 - accuracy: 0.0043 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2780 - val_accuracy: 0.0043 - val_top_k_categorical_accuracy: 0.0061 - lr: 5.0000e-04
Epoch 20/20
405/408 [============================>.] - ETA: 0s - loss: 5.2787 - accuracy: 0.0025 - top_k_categorical_accuracy: 0.0000e+00     
Epoch 20: val_loss improved from 5.27799 to 5.27799, saving model to googlenet_finetuned_best.keras
408/408 [==============================] - 14s 35ms/step - loss: 5.2790 - accuracy: 0.0025 - top_k_categorical_accuracy: 0.0000e+00 - val_loss: 5.2780 - val_accuracy: 0.0037 - val_top_k_categorical_accuracy: 0.0061 - lr: 5.0000e-04
Restoring model weights from the end of the best epoch: 20.

Train Val Loss Graph

In [97]:
plot_training_history(history)
In [98]:
# Get all predictions and true labels
y_true = []
y_pred = []

for X_batch, y_batch in val_ds:
    preds = googlenet_model_tuned.predict(X_batch)
    y_pred_batch = np.argmax(preds, axis=1)
    y_true_batch = y_batch.numpy() if hasattr(y_batch, "numpy") else y_batch
    y_true.extend(y_true_batch)
    y_pred.extend(y_pred_batch)

1/1 [==============================] - 7s 7s/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 162ms/step
1/1 [==============================] - 0s 31ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 30ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 29ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 30ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 30ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 30ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 30ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 30ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 30ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 31ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 28ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 30ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 31ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
1/1 [==============================] - 0s 27ms/step
In [99]:
target_names = label_encoder.classes_ if 'label_encoder' in globals() else None
In [100]:
report = classification_report(
    y_true, y_pred,
    target_names=target_names,
    output_dict=True,
    zero_division=1
)
df_googlenet_tuned_report = pd.DataFrame(report).transpose()
acc = accuracy_score(y_true, y_pred)
df_googlenet_tuned_report.loc["overall_accuracy"] = [acc, None, None, None]
df_googlenet_tuned_report.to_csv("googlenet_tuned_classification_report.csv")
print(f"Tuned GoogLeNet Accuracy: {acc:.4f}")
print("Average Summary Metrics:")
print(df_googlenet_tuned_report.tail(3)[["precision", "recall", "f1-score"]])

Tuned GoogLeNet Accuracy: 0.0037
Average Summary Metrics:
                  precision    recall  f1-score
macro avg          0.638079  0.002648  0.000545
weighted avg       0.628435  0.003683  0.000747
overall_accuracy   0.003683       NaN       NaN

confusion matrix for tuned

In [101]:
cm = confusion_matrix(y_true, y_pred)
In [102]:
df_support = df_googlenet_tuned_report.iloc[:-3]
top_10_classes = df_support.sort_values("support", ascending=False).head(10).index.tolist()
In [103]:
# Get class indices (map from class name to index)
if target_names is not None:
    top_10_indices = [np.where(target_names == cls)[0][0] for cls in top_10_classes]
else:
    top_10_indices = list(map(int, top_10_classes))  # fallback if no class names
In [104]:
cm_top10 = cm[np.ix_(top_10_indices, top_10_indices)]
In [105]:
# Plot
plt.figure(figsize=(10, 8))
sns.heatmap(cm_top10, annot=True, fmt='d',
            xticklabels=top_10_classes,
            yticklabels=top_10_classes,
            cmap='Blues')
plt.title("Tuned GoogLeNet - Confusion Matrix (Top 10 Classes)")
plt.xlabel("Predicted")
plt.ylabel("True")
plt.tight_layout()
plt.show()

GoogleNet FineTuned Summary¶

  • The untuned GoogLeNet model achieved an accuracy of 0.49%, slightly outperforming the tuned model, which dropped to 0.43%.
  • Although macro and weighted precision appear high, the recall and F1-scores are nearly zero, confirming that the model rarely makes correct predictions.
  • This indicates that tuning did not improve performance and may have disrupted learning, likely due to preprocessing inconsistencies or label mapping mismatches or fine-tuning strategies

Fine Tuning Of ResNet Model¶

for Classimbalance

In [106]:
#encoding labels
label_encoder = LabelEncoder()
df_training['labels_encoded'] = label_encoder.fit_transform(df_training['labels'])
In [107]:
df_training['labels'] = df_training['labels'].astype(str)
df_val['labels'] = df_val['labels'].astype(str)
In [108]:
# Calculate class weights (important for class imbalance)
class_weights = class_weight.compute_class_weight(
    'balanced', 
    classes=np.unique(df_training['labels_encoded']),
    y=df_training['labels_encoded']
)
class_weights = dict(enumerate(class_weights))  # Convert to dictionary
In [109]:
batch_size = 32
image_size = (224, 224)
In [110]:
train_datagen = ImageDataGenerator(
    rotation_range=20,
    width_shift_range=0.2,
    height_shift_range=0.2,
    shear_range=0.2,
    zoom_range=0.2,
    horizontal_flip=True,
    fill_mode='nearest'
    #,preprocessing_function=resnet_preprocess
)

train_generator = train_datagen.flow_from_dataframe(
    df_training,
    x_col='Image_Path',
    y_col='labels',
    target_size=image_size,
    batch_size=batch_size,
    class_mode='categorical'
)

Found 8144 validated image filenames belonging to 196 classes.
In [111]:
#val_datagen = ImageDataGenerator(preprocessing_function=resnet_preprocess)
val_datagen = ImageDataGenerator()

val_generator = val_datagen.flow_from_dataframe(
    df_val,
    x_col='Image_Path',
    y_col='labels',
    target_size=image_size,
    batch_size=batch_size,
    class_mode='categorical'
)

Found 1629 validated image filenames belonging to 196 classes.

model definition

In [112]:
# Load ResNet50 base model without the top layer
base_model = ResNet50(weights='imagenet', include_top=False, input_shape=(224, 224, 3))

# Unfreeze some layers of the base model for fine-tuning (last 40 layers)
for layer in base_model.layers:
    layer.trainable = False
    
for layer in base_model.layers[-40:]:
    layer.trainable = True

num_classes = df_train['labels'].nunique()

# Add custom classification layers
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(512, activation='relu')(x)
x = Dropout(0.5)(x)
x = Dense(num_classes, activation='softmax')(x)  # Output layer

# Define model
resnet_tuned_model = Model(inputs=base_model.input, outputs=x)

# Compile the model again with a higher learning rate
resnet_tuned_model.compile(optimizer=Adam(learning_rate=1e-5), loss='categorical_crossentropy', metrics=['accuracy'])
In [113]:
resnet_tuned_model.summary()

Model: "model_4"
__________________________________________________________________________________________________
 Layer (type)                Output Shape                 Param #   Connected to                  
==================================================================================================
 input_5 (InputLayer)        [(None, 224, 224, 3)]        0         []                            
                                                                                                  
 conv1_pad (ZeroPadding2D)   (None, 230, 230, 3)          0         ['input_5[0][0]']             
                                                                                                  
 conv1_conv (Conv2D)         (None, 112, 112, 64)         9472      ['conv1_pad[0][0]']           
                                                                                                  
 conv1_bn (BatchNormalizati  (None, 112, 112, 64)         256       ['conv1_conv[0][0]']          
 on)                                                                                              
                                                                                                  
 conv1_relu (Activation)     (None, 112, 112, 64)         0         ['conv1_bn[0][0]']            
                                                                                                  
 pool1_pad (ZeroPadding2D)   (None, 114, 114, 64)         0         ['conv1_relu[0][0]']          
                                                                                                  
 pool1_pool (MaxPooling2D)   (None, 56, 56, 64)           0         ['pool1_pad[0][0]']           
                                                                                                  
 conv2_block1_1_conv (Conv2  (None, 56, 56, 64)           4160      ['pool1_pool[0][0]']          
 D)                                                                                               
                                                                                                  
 conv2_block1_1_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block1_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block1_1_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block1_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block1_2_conv (Conv2  (None, 56, 56, 64)           36928     ['conv2_block1_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block1_2_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block1_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block1_2_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block1_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block1_0_conv (Conv2  (None, 56, 56, 256)          16640     ['pool1_pool[0][0]']          
 D)                                                                                               
                                                                                                  
 conv2_block1_3_conv (Conv2  (None, 56, 56, 256)          16640     ['conv2_block1_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block1_0_bn (BatchNo  (None, 56, 56, 256)          1024      ['conv2_block1_0_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block1_3_bn (BatchNo  (None, 56, 56, 256)          1024      ['conv2_block1_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block1_add (Add)      (None, 56, 56, 256)          0         ['conv2_block1_0_bn[0][0]',   
                                                                     'conv2_block1_3_bn[0][0]']   
                                                                                                  
 conv2_block1_out (Activati  (None, 56, 56, 256)          0         ['conv2_block1_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv2_block2_1_conv (Conv2  (None, 56, 56, 64)           16448     ['conv2_block1_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv2_block2_1_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block2_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block2_1_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block2_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block2_2_conv (Conv2  (None, 56, 56, 64)           36928     ['conv2_block2_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block2_2_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block2_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block2_2_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block2_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block2_3_conv (Conv2  (None, 56, 56, 256)          16640     ['conv2_block2_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block2_3_bn (BatchNo  (None, 56, 56, 256)          1024      ['conv2_block2_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block2_add (Add)      (None, 56, 56, 256)          0         ['conv2_block1_out[0][0]',    
                                                                     'conv2_block2_3_bn[0][0]']   
                                                                                                  
 conv2_block2_out (Activati  (None, 56, 56, 256)          0         ['conv2_block2_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv2_block3_1_conv (Conv2  (None, 56, 56, 64)           16448     ['conv2_block2_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv2_block3_1_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block3_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block3_1_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block3_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block3_2_conv (Conv2  (None, 56, 56, 64)           36928     ['conv2_block3_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block3_2_bn (BatchNo  (None, 56, 56, 64)           256       ['conv2_block3_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block3_2_relu (Activ  (None, 56, 56, 64)           0         ['conv2_block3_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv2_block3_3_conv (Conv2  (None, 56, 56, 256)          16640     ['conv2_block3_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv2_block3_3_bn (BatchNo  (None, 56, 56, 256)          1024      ['conv2_block3_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv2_block3_add (Add)      (None, 56, 56, 256)          0         ['conv2_block2_out[0][0]',    
                                                                     'conv2_block3_3_bn[0][0]']   
                                                                                                  
 conv2_block3_out (Activati  (None, 56, 56, 256)          0         ['conv2_block3_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv3_block1_1_conv (Conv2  (None, 28, 28, 128)          32896     ['conv2_block3_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv3_block1_1_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block1_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block1_1_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block1_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block1_2_conv (Conv2  (None, 28, 28, 128)          147584    ['conv3_block1_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block1_2_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block1_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block1_2_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block1_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block1_0_conv (Conv2  (None, 28, 28, 512)          131584    ['conv2_block3_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv3_block1_3_conv (Conv2  (None, 28, 28, 512)          66048     ['conv3_block1_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block1_0_bn (BatchNo  (None, 28, 28, 512)          2048      ['conv3_block1_0_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block1_3_bn (BatchNo  (None, 28, 28, 512)          2048      ['conv3_block1_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block1_add (Add)      (None, 28, 28, 512)          0         ['conv3_block1_0_bn[0][0]',   
                                                                     'conv3_block1_3_bn[0][0]']   
                                                                                                  
 conv3_block1_out (Activati  (None, 28, 28, 512)          0         ['conv3_block1_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv3_block2_1_conv (Conv2  (None, 28, 28, 128)          65664     ['conv3_block1_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv3_block2_1_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block2_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block2_1_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block2_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block2_2_conv (Conv2  (None, 28, 28, 128)          147584    ['conv3_block2_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block2_2_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block2_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block2_2_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block2_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block2_3_conv (Conv2  (None, 28, 28, 512)          66048     ['conv3_block2_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block2_3_bn (BatchNo  (None, 28, 28, 512)          2048      ['conv3_block2_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block2_add (Add)      (None, 28, 28, 512)          0         ['conv3_block1_out[0][0]',    
                                                                     'conv3_block2_3_bn[0][0]']   
                                                                                                  
 conv3_block2_out (Activati  (None, 28, 28, 512)          0         ['conv3_block2_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv3_block3_1_conv (Conv2  (None, 28, 28, 128)          65664     ['conv3_block2_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv3_block3_1_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block3_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block3_1_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block3_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block3_2_conv (Conv2  (None, 28, 28, 128)          147584    ['conv3_block3_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block3_2_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block3_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block3_2_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block3_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block3_3_conv (Conv2  (None, 28, 28, 512)          66048     ['conv3_block3_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block3_3_bn (BatchNo  (None, 28, 28, 512)          2048      ['conv3_block3_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block3_add (Add)      (None, 28, 28, 512)          0         ['conv3_block2_out[0][0]',    
                                                                     'conv3_block3_3_bn[0][0]']   
                                                                                                  
 conv3_block3_out (Activati  (None, 28, 28, 512)          0         ['conv3_block3_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv3_block4_1_conv (Conv2  (None, 28, 28, 128)          65664     ['conv3_block3_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv3_block4_1_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block4_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block4_1_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block4_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block4_2_conv (Conv2  (None, 28, 28, 128)          147584    ['conv3_block4_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block4_2_bn (BatchNo  (None, 28, 28, 128)          512       ['conv3_block4_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block4_2_relu (Activ  (None, 28, 28, 128)          0         ['conv3_block4_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv3_block4_3_conv (Conv2  (None, 28, 28, 512)          66048     ['conv3_block4_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv3_block4_3_bn (BatchNo  (None, 28, 28, 512)          2048      ['conv3_block4_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv3_block4_add (Add)      (None, 28, 28, 512)          0         ['conv3_block3_out[0][0]',    
                                                                     'conv3_block4_3_bn[0][0]']   
                                                                                                  
 conv3_block4_out (Activati  (None, 28, 28, 512)          0         ['conv3_block4_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block1_1_conv (Conv2  (None, 14, 14, 256)          131328    ['conv3_block4_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block1_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block1_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block1_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block1_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block1_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block1_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block1_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block1_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block1_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block1_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block1_0_conv (Conv2  (None, 14, 14, 1024)         525312    ['conv3_block4_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block1_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block1_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block1_0_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block1_0_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block1_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block1_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block1_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block1_0_bn[0][0]',   
                                                                     'conv4_block1_3_bn[0][0]']   
                                                                                                  
 conv4_block1_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block1_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block2_1_conv (Conv2  (None, 14, 14, 256)          262400    ['conv4_block1_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block2_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block2_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block2_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block2_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block2_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block2_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block2_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block2_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block2_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block2_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block2_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block2_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block2_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block2_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block2_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block1_out[0][0]',    
                                                                     'conv4_block2_3_bn[0][0]']   
                                                                                                  
 conv4_block2_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block2_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block3_1_conv (Conv2  (None, 14, 14, 256)          262400    ['conv4_block2_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block3_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block3_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block3_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block3_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block3_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block3_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block3_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block3_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block3_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block3_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block3_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block3_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block3_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block3_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block3_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block2_out[0][0]',    
                                                                     'conv4_block3_3_bn[0][0]']   
                                                                                                  
 conv4_block3_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block3_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block4_1_conv (Conv2  (None, 14, 14, 256)          262400    ['conv4_block3_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block4_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block4_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block4_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block4_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block4_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block4_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block4_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block4_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block4_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block4_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block4_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block4_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block4_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block4_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block4_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block3_out[0][0]',    
                                                                     'conv4_block4_3_bn[0][0]']   
                                                                                                  
 conv4_block4_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block4_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block5_1_conv (Conv2  (None, 14, 14, 256)          262400    ['conv4_block4_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block5_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block5_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block5_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block5_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block5_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block5_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block5_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block5_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block5_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block5_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block5_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block5_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block5_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block5_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block5_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block4_out[0][0]',    
                                                                     'conv4_block5_3_bn[0][0]']   
                                                                                                  
 conv4_block5_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block5_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv4_block6_1_conv (Conv2  (None, 14, 14, 256)          262400    ['conv4_block5_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv4_block6_1_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block6_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block6_1_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block6_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block6_2_conv (Conv2  (None, 14, 14, 256)          590080    ['conv4_block6_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block6_2_bn (BatchNo  (None, 14, 14, 256)          1024      ['conv4_block6_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block6_2_relu (Activ  (None, 14, 14, 256)          0         ['conv4_block6_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv4_block6_3_conv (Conv2  (None, 14, 14, 1024)         263168    ['conv4_block6_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv4_block6_3_bn (BatchNo  (None, 14, 14, 1024)         4096      ['conv4_block6_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv4_block6_add (Add)      (None, 14, 14, 1024)         0         ['conv4_block5_out[0][0]',    
                                                                     'conv4_block6_3_bn[0][0]']   
                                                                                                  
 conv4_block6_out (Activati  (None, 14, 14, 1024)         0         ['conv4_block6_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv5_block1_1_conv (Conv2  (None, 7, 7, 512)            524800    ['conv4_block6_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv5_block1_1_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block1_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block1_1_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block1_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block1_2_conv (Conv2  (None, 7, 7, 512)            2359808   ['conv5_block1_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block1_2_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block1_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block1_2_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block1_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block1_0_conv (Conv2  (None, 7, 7, 2048)           2099200   ['conv4_block6_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv5_block1_3_conv (Conv2  (None, 7, 7, 2048)           1050624   ['conv5_block1_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block1_0_bn (BatchNo  (None, 7, 7, 2048)           8192      ['conv5_block1_0_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block1_3_bn (BatchNo  (None, 7, 7, 2048)           8192      ['conv5_block1_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block1_add (Add)      (None, 7, 7, 2048)           0         ['conv5_block1_0_bn[0][0]',   
                                                                     'conv5_block1_3_bn[0][0]']   
                                                                                                  
 conv5_block1_out (Activati  (None, 7, 7, 2048)           0         ['conv5_block1_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv5_block2_1_conv (Conv2  (None, 7, 7, 512)            1049088   ['conv5_block1_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv5_block2_1_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block2_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block2_1_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block2_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block2_2_conv (Conv2  (None, 7, 7, 512)            2359808   ['conv5_block2_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block2_2_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block2_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block2_2_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block2_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block2_3_conv (Conv2  (None, 7, 7, 2048)           1050624   ['conv5_block2_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block2_3_bn (BatchNo  (None, 7, 7, 2048)           8192      ['conv5_block2_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block2_add (Add)      (None, 7, 7, 2048)           0         ['conv5_block1_out[0][0]',    
                                                                     'conv5_block2_3_bn[0][0]']   
                                                                                                  
 conv5_block2_out (Activati  (None, 7, 7, 2048)           0         ['conv5_block2_add[0][0]']    
 on)                                                                                              
                                                                                                  
 conv5_block3_1_conv (Conv2  (None, 7, 7, 512)            1049088   ['conv5_block2_out[0][0]']    
 D)                                                                                               
                                                                                                  
 conv5_block3_1_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block3_1_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block3_1_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block3_1_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block3_2_conv (Conv2  (None, 7, 7, 512)            2359808   ['conv5_block3_1_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block3_2_bn (BatchNo  (None, 7, 7, 512)            2048      ['conv5_block3_2_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block3_2_relu (Activ  (None, 7, 7, 512)            0         ['conv5_block3_2_bn[0][0]']   
 ation)                                                                                           
                                                                                                  
 conv5_block3_3_conv (Conv2  (None, 7, 7, 2048)           1050624   ['conv5_block3_2_relu[0][0]'] 
 D)                                                                                               
                                                                                                  
 conv5_block3_3_bn (BatchNo  (None, 7, 7, 2048)           8192      ['conv5_block3_3_conv[0][0]'] 
 rmalization)                                                                                     
                                                                                                  
 conv5_block3_add (Add)      (None, 7, 7, 2048)           0         ['conv5_block2_out[0][0]',    
                                                                     'conv5_block3_3_bn[0][0]']   
                                                                                                  
 conv5_block3_out (Activati  (None, 7, 7, 2048)           0         ['conv5_block3_add[0][0]']    
 on)                                                                                              
                                                                                                  
 global_average_pooling2d_4  (None, 2048)                 0         ['conv5_block3_out[0][0]']    
  (GlobalAveragePooling2D)                                                                        
                                                                                                  
 dense_11 (Dense)            (None, 512)                  1049088   ['global_average_pooling2d_4[0
                                                                    ][0]']                        
                                                                                                  
 dropout_5 (Dropout)         (None, 512)                  0         ['dense_11[0][0]']            
                                                                                                  
 dense_12 (Dense)            (None, 196)                  100548    ['dropout_5[0][0]']           
                                                                                                  
==================================================================================================
Total params: 24737348 (94.37 MB)
Trainable params: 16981444 (64.78 MB)
Non-trainable params: 7755904 (29.59 MB)
__________________________________________________________________________________________________

Data Generation

In [114]:
# Custom data generator
""" def custom_data_generator(generator, class_weights):
    while True:
        x, y = next(generator)
        sample_weights = np.array([class_weights[np.argmax(label)] for label in y])
        yield x, y, sample_weights """

def custom_data_generator(generator, class_weights):
    while True:
        x, y = next(generator)
        sample_weights = np.array([class_weights[label] for label in np.argmax(y, axis=1)])
        yield x, y, sample_weights
In [115]:
# Define data augmentation
#train_datagen = ImageDataGenerator(
#    rotation_range=20,
#    width_shift_range=0.2,
#    height_shift_range=0.2,
#    shear_range=0.2,
#    zoom_range=0.2,
#    horizontal_flip=True,
#    fill_mode='nearest',
#    preprocessing_function=preprocess_input  # Use the ResNet50 preprocessing function
#)


# Load data using flow_from_dataframe
#print(df_train.columns)  # Check the columns in the DataFrame

# Convert labels to string format if they are not already
#df_train['labels'] = df_train['labels'].astype(str)
In [116]:
# Create training data generator
#train_generator = train_datagen.flow_from_dataframe(
#    df_train,
#    x_col='Image_Path',
#    y_col='labels',  # Now it should be in the correct format
#    target_size=(224, 224),
#    batch_size=batch_size,
#    class_mode='categorical'
#)
In [117]:
# Create validation data generator (assuming df_val is defined)
#val_datagen = ImageDataGenerator(preprocessing_function=preprocess_input)  # No augmentation for validation
#val_generator = val_datagen.flow_from_dataframe(
#    df_val,
#    x_col='Image_Path',
#    y_col='labels',
#    target_size=(224, 224),
#    batch_size=batch_size,
#    class_mode='categorical'
#)
In [118]:
# Train the model for a few epochs
#steps_per_epoch = len(df_training) // batch_size
#validation_steps = len(df_val) // batch_size
steps_per_epoch = np.ceil(len(df_training) / batch_size).astype(int)
validation_steps = np.ceil(len(df_val) / batch_size).astype(int)


# Define number of fine-tuning epochs
fine_tune_epochs = 20

# Define callbacks
early_stopping = EarlyStopping(monitor='val_loss', patience=5, restore_best_weights=True)
model_checkpoint = ModelCheckpoint('best_model_tuned_resnet.keras', save_best_only=True, monitor='val_loss')
In [119]:
# Train the model with callbacks and custom data generator
#resnet_history_fine_tune = resnet_model.fit(
#    custom_data_generator(train_generator, class_weights),
#    steps_per_epoch=steps_per_epoch,
#    validation_data=val_generator,
#    validation_steps=validation_steps,
#    epochs=fine_tune_epochs,
#    callbacks=[early_stopping, model_checkpoint]
#)
resnet_history_fine_tune = resnet_tuned_model.fit(
    train_generator,
    steps_per_epoch=steps_per_epoch,
    validation_data=val_generator,
    validation_steps=validation_steps,
    epochs=20,
    callbacks=[early_stopping, model_checkpoint]
    ,class_weight=class_weights
)

Epoch 1/20
255/255 [==============================] - 104s 348ms/step - loss: 5.5599 - accuracy: 0.0054 - val_loss: 5.2763 - val_accuracy: 0.0055
Epoch 2/20
255/255 [==============================] - 86s 339ms/step - loss: 5.3193 - accuracy: 0.0077 - val_loss: 5.1636 - val_accuracy: 0.0135
Epoch 3/20
255/255 [==============================] - 87s 339ms/step - loss: 5.2150 - accuracy: 0.0138 - val_loss: 5.0478 - val_accuracy: 0.0344
Epoch 4/20
255/255 [==============================] - 86s 337ms/step - loss: 5.1253 - accuracy: 0.0249 - val_loss: 4.9025 - val_accuracy: 0.0552
Epoch 5/20
255/255 [==============================] - 86s 339ms/step - loss: 5.0139 - accuracy: 0.0332 - val_loss: 4.7125 - val_accuracy: 0.0896
Epoch 6/20
255/255 [==============================] - 87s 339ms/step - loss: 4.8937 - accuracy: 0.0438 - val_loss: 4.4970 - val_accuracy: 0.1295
Epoch 7/20
255/255 [==============================] - 86s 339ms/step - loss: 4.7345 - accuracy: 0.0667 - val_loss: 4.2544 - val_accuracy: 0.1878
Epoch 8/20
255/255 [==============================] - 87s 340ms/step - loss: 4.5798 - accuracy: 0.0829 - val_loss: 4.0057 - val_accuracy: 0.2400
Epoch 9/20
255/255 [==============================] - 86s 338ms/step - loss: 4.3906 - accuracy: 0.1106 - val_loss: 3.7407 - val_accuracy: 0.2775
Epoch 10/20
255/255 [==============================] - 86s 338ms/step - loss: 4.2036 - accuracy: 0.1342 - val_loss: 3.4715 - val_accuracy: 0.3217
Epoch 11/20
255/255 [==============================] - 87s 339ms/step - loss: 4.0151 - accuracy: 0.1580 - val_loss: 3.2367 - val_accuracy: 0.3671
Epoch 12/20
255/255 [==============================] - 86s 338ms/step - loss: 3.7990 - accuracy: 0.1933 - val_loss: 3.0037 - val_accuracy: 0.4033
Epoch 13/20
255/255 [==============================] - 86s 339ms/step - loss: 3.6676 - accuracy: 0.2101 - val_loss: 2.8150 - val_accuracy: 0.4432
Epoch 14/20
255/255 [==============================] - 87s 340ms/step - loss: 3.4788 - accuracy: 0.2410 - val_loss: 2.5960 - val_accuracy: 0.4948
Epoch 15/20
255/255 [==============================] - 86s 338ms/step - loss: 3.3210 - accuracy: 0.2653 - val_loss: 2.4155 - val_accuracy: 0.5285
Epoch 16/20
255/255 [==============================] - 87s 340ms/step - loss: 3.1629 - accuracy: 0.2942 - val_loss: 2.2691 - val_accuracy: 0.5654
Epoch 17/20
255/255 [==============================] - 86s 337ms/step - loss: 3.0306 - accuracy: 0.3116 - val_loss: 2.1205 - val_accuracy: 0.5838
Epoch 18/20
255/255 [==============================] - 86s 336ms/step - loss: 2.8561 - accuracy: 0.3466 - val_loss: 1.9307 - val_accuracy: 0.6120
Epoch 19/20
255/255 [==============================] - 86s 337ms/step - loss: 2.7541 - accuracy: 0.3642 - val_loss: 1.7872 - val_accuracy: 0.6568
Epoch 20/20
255/255 [==============================] - 86s 339ms/step - loss: 2.5899 - accuracy: 0.4015 - val_loss: 1.6911 - val_accuracy: 0.6777
In [120]:
plot_training_history(resnet_history_fine_tune)
In [121]:
num_samples = len(df_val)
X_val = np.array(df_val['image'].tolist()).astype(np.float32)  # Keep as list to avoid memory burst
 
y_val_true = np.array([np.argmax(label) for label in df_val['label_categorical']])
y_val_pred = []

for i in range(0, num_samples, batch_size):
    batch_imgs = X_val[i:i+batch_size]     
    preds = resnet_tuned_model.predict(batch_imgs, verbose=0)
    batch_preds = np.argmax(preds, axis=1)
    y_val_pred.extend(batch_preds)

y_val_pred = np.array(y_val_pred)
In [122]:
print("X_val shape:", X_val.shape)
print("X_val dtype:", X_val.dtype)

X_val shape: (1629, 224, 224, 3)
X_val dtype: float32
In [123]:
target_names = label_encoder.classes_ if 'label_encoder' in globals() else None
In [124]:
resnet_tuned_report = classification_report(
    y_val_true, y_val_pred,
    target_names=target_names,
    output_dict=True,
    zero_division=1  # Avoid divide-by-zero errors
)
In [125]:
#print("Unique y_true:", np.unique(y_val_true))
#print("Unique y_pred:", np.unique(y_val_pred))
In [126]:
#unique_preds, counts = np.unique(y_val_pred, return_counts=True)
#print("Predicted class distribution:", dict(zip(unique_preds, counts)))
In [127]:
df_resnet_tuned_report = pd.DataFrame(resnet_tuned_report).transpose()
In [128]:
acc = accuracy_score(y_val_true, y_val_pred)
df_resnet_tuned_report.loc["overall_accuracy"] = [acc, None, None, None]
In [129]:
df_resnet_tuned_report.to_csv("resnet_tuned_classification_report.csv")
In [130]:
print(f"Tuned ResNet Accuracy: {acc:.4f}")
print("Average Resnet Summary Metrics:")
print(df_resnet_tuned_report.tail(3)[["precision", "recall", "f1-score"]])

Tuned ResNet Accuracy: 0.0037
Average Resnet Summary Metrics:
                  precision    recall  f1-score
macro avg          0.974528  0.005782  0.000076
weighted avg       0.975470  0.003683  0.000050
overall_accuracy   0.003683       NaN       NaN
In [131]:
cm = confusion_matrix(y_true, y_pred)
In [132]:
df_support = df_resnet_tuned_report.iloc[:-3]
top_10_classes = df_support.sort_values("support", ascending=False).head(10).index.tolist()
In [133]:
if target_names is not None:
    top_10_indices = [np.where(target_names == cls)[0][0] for cls in top_10_classes]
else:
    top_10_indices = list(map(int, top_10_classes))  # fallback if no class names
	
cm_top10 = cm[np.ix_(top_10_indices, top_10_indices)]
In [134]:
plt.figure(figsize=(10, 8))
sns.heatmap(cm_top10, annot=True, fmt='d',
            xticklabels=top_10_classes,
            yticklabels=top_10_classes,
            cmap='Blues')
plt.title("Resnet Tuned - Confusion Matrix (Top 10 Classes)")
plt.xlabel("Predicted")
plt.ylabel("True")
plt.tight_layout()
plt.show()

ResNet FineTuned Summary¶

  • Despite tuning, the ResNet model's performance declined, with accuracy dropping from 0.55% (untuned) to 0.43% (tuned).
  • Precision remains misleadingly high due to sparse predictions, while recall and F1-scores are nearly zero in both cases — indicating that the model fails to generalize across classes.
  • Which suggests issues such as label mismatch, improper preprocessing, or training data imbalance, leading to severe underfitting or poor class prediction confidence even after tuning.-

Model Decision¶

In [135]:
# First, add a column to identify each model
df_resnet_classification_report_tail = df_resnet_classification_report.tail(4).copy()
df_resnet_classification_report_tail['Model'] = 'ResNet Untuned (10 Epochs)'

df_resnet_tuned_report_tail = df_resnet_tuned_report.tail(4).copy()
df_resnet_tuned_report_tail['Model'] = 'ResNet Tuned (20 Epochs)'


df_googlenet_classification_report_tail = df_googlenet_classification_report.tail(4).copy()
df_googlenet_classification_report_tail['Model'] = 'GoogLeNet Untuned (10 Epochs)'

df_googlenet_tuned_report_tail = df_googlenet_tuned_report.tail(4).copy()
df_googlenet_tuned_report_tail['Model'] = 'GoogLeNet Tuned (20 Epochs)'
In [136]:
df_combined_tail = pd.concat([
    df_resnet_classification_report_tail,
    df_resnet_tuned_report_tail,
    df_googlenet_classification_report_tail,
    df_googlenet_tuned_report_tail
])
In [137]:
df_combined_tail = df_combined_tail.reset_index().rename(columns={'index': 'Metric'})
In [138]:
df_combined_tail = df_combined_tail[['Model', 'Metric', 'precision', 'recall', 'f1-score']]
In [139]:
df_combined_tail.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 16 entries, 0 to 15
Data columns (total 5 columns):
 #   Column     Non-Null Count  Dtype  
---  ------     --------------  -----  
 0   Model      16 non-null     object 
 1   Metric     16 non-null     object 
 2   precision  16 non-null     float64
 3   recall     12 non-null     float64
 4   f1-score   12 non-null     float64
dtypes: float64(3), object(2)
memory usage: 768.0+ bytes
In [140]:
df_combined_tail
Out[140]:
Model Metric precision recall f1-score
0 ResNet Untuned (10 Epochs) accuracy 0.010436 0.010436 0.010436
1 ResNet Untuned (10 Epochs) macro avg 0.989922 0.007070 0.000247
2 ResNet Untuned (10 Epochs) weighted avg 0.985440 0.010436 0.000340
3 ResNet Untuned (10 Epochs) overall_accuracy 0.010436 NaN NaN
4 ResNet Tuned (20 Epochs) accuracy 0.003683 0.003683 0.003683
5 ResNet Tuned (20 Epochs) macro avg 0.974528 0.005782 0.000076
6 ResNet Tuned (20 Epochs) weighted avg 0.975470 0.003683 0.000050
7 ResNet Tuned (20 Epochs) overall_accuracy 0.003683 NaN NaN
8 GoogLeNet Untuned (10 Epochs) accuracy 0.240638 0.240638 0.240638
9 GoogLeNet Untuned (10 Epochs) macro avg 0.357037 0.240374 0.219307
10 GoogLeNet Untuned (10 Epochs) weighted avg 0.384768 0.240638 0.231432
11 GoogLeNet Untuned (10 Epochs) overall_accuracy 0.240638 NaN NaN
12 GoogLeNet Tuned (20 Epochs) accuracy 0.003683 0.003683 0.003683
13 GoogLeNet Tuned (20 Epochs) macro avg 0.638079 0.002648 0.000545
14 GoogLeNet Tuned (20 Epochs) weighted avg 0.628435 0.003683 0.000747
15 GoogLeNet Tuned (20 Epochs) overall_accuracy 0.003683 NaN NaN

Model Selection¶

  • Among all models tested, GoogLeNet (10 Epochs) showed the best performance, achieving an accuracy of 22.9% and a macro-average F1-score of 0.2056, significantly outperforming all other models.
  • Despite additional training and tuning, both ResNet and GoogLeNet tuned models failed to generalize

hence the Final Model Selected for Test Evaluation:GoogLeNet Untuned Version with 10 Epohs

Untuned GoogleNet Model Validated Against Test Data Set¶

In [141]:
batch_size=16
# Step 1: Get true labels
y_test_true = np.array([np.argmax(label) for label in df_testing['label_categorical']])

# Step 2: Predict in batches
y_test_pred = []

for i in range(0, len(df_testing), batch_size):
    batch_imgs = np.array(df_testing['image'].tolist()[i:i+batch_size])
    preds = googlenet_model.predict(batch_imgs, verbose=0)
    batch_preds = np.argmax(preds, axis=1)
    y_test_pred.extend(batch_preds)

y_test_pred = np.array(y_test_pred)
In [142]:
target_names = label_encoder.classes_ if 'label_encoder' in globals() else None
In [143]:
final_googlenet_untuned_report = classification_report(
    y_test_true, y_test_pred,
    target_names=target_names,
    output_dict=True,
    zero_division=1  # Avoid divide-by-zero errors
)
In [144]:
df_final_googlenet_untuned_report = pd.DataFrame(final_googlenet_untuned_report).transpose()
In [145]:
acc = accuracy_score(y_test_true, y_test_pred)
In [146]:
df_final_googlenet_untuned_report.loc["overall_accuracy"]= [acc, None, None, None]
In [147]:
df_final_googlenet_untuned_report.to_csv("df_final_googlenet_untuned_classification_report.csv")
In [148]:
print(f"Final GoogleNet(Untuned) against test data Accuracy: {acc:.4f}")
print("Final Untuned GoogleNet metrics against test:")
print(df_final_googlenet_untuned_report.tail(3)[["precision", "recall", "f1-score"]])

Final GoogleNet(Untuned) against test data Accuracy: 0.2406
Final Untuned GoogleNet metrics against test:
                  precision    recall  f1-score
macro avg          0.340414  0.239268  0.229648
weighted avg       0.340646  0.240642  0.229778
overall_accuracy   0.240642       NaN       NaN
In [149]:
cm = confusion_matrix(y_true, y_pred)
df_support = df_final_googlenet_untuned_report.iloc[:-3]
top_10_classes = df_support.sort_values("support", ascending=False).head(10).index.tolist()
In [150]:
if target_names is not None:
    top_10_indices = [np.where(target_names == cls)[0][0] for cls in top_10_classes]
else:
    top_10_indices = list(map(int, top_10_classes))  # fallback if no class names
In [151]:
cm_top10 = cm[np.ix_(top_10_indices, top_10_indices)]
In [152]:
plt.figure(figsize=(10, 8))
sns.heatmap(cm_top10, annot=True, fmt='d',
            xticklabels=top_10_classes,
            yticklabels=top_10_classes,
            cmap='Blues')
plt.title("Tuned GoogLeNet - Confusion Matrix (Top 10 Classes)")
plt.xlabel("Predicted")
plt.ylabel("True")
plt.tight_layout()
plt.show()

Final Outcome (Test Data Evaluation)¶

  • The untuned GoogLeNet model achieved a test accuracy of 21.94%, closely aligning with its validation performance (~22.90%).
  • The macro and weighted average F1-scores (~0.205) indicate a reasonable baseline performance across 196 fine-grained car classes.
  • This confirms the model has learned meaningful patterns and generalizes well, though this model requires further tuning/training/restructuring would be required
*The GoogleNet is a pure classification model and does not support localization or masking of car regions in images*¶
*Since RCNN and its hybrids require a base model capable of region proposals or feature maps for bounding box regression and segmentation*¶
*As the current GoogLeNet model is not suitable for region-based masking required in RCNN workflows, we will move forward with implementing YOLO for object detection and localization.*¶
In [ ]: